Review

ChemBioChem doi.org/10.1002/cbic.202300120

www.chembiochem.org

Quantum Computing for Molecular Biology**

Alberto Baiardi,*[a] Matthias Christandl,*[b] and Markus Reiher*[a]

Molecular biology and biochemistry interpret microscopic tages; however, this comes at the cost that all quantum
processes in the living world in terms of molecular structures correlations and the rigorous many-particle nature of the
and their interactions, which are quantum mechanical by their interactions are omitted. In this work, we discuss how quantum
very nature. Whereas the theoretical foundations of these computation may advance the practical usefulness of the
interactions are well established, the computational solution of quantum foundations of molecular biology by offering compu-
the relevant quantum mechanical equations is very hard. tational advantages for simulations of biomolecules. We not
However, much of molecular function in biology can be only discuss typical quantum mechanical problems of the
understood in terms of classical mechanics, where the inter- electronic structure of biomolecules in this context, but also
actions of electrons and nuclei have been mapped onto consider the dominating classical problems (such as protein
effective classical surrogate potentials that model the inter- folding and drug design) as well as data-driven approaches of
action of atoms or even larger entities. The simple mathematical bioinformatics and the degree to which they might become
structure of these potentials offers huge computational advan- amenable to quantum simulation and quantum computation.

1. Introduction sufficient in order to create functional networks of biochemical

pathways (classical examples are the glycolysis pathway and
The elucidation of the molecular foundations of biology has the Krebs cycle), leading to an understanding of molecular
changed the way we understand biology at the cellular level. functions that may be viewed as a what-comes-next static
Examples are discoveries in the molecular machinery of cells picture of molecular events.
such as the structural characterization of proteins and their Still, only detailed quantitative physical simulations – in
mode of action at atomistic resolution[1–4] as well as the genome parallel to detailed experiments with a high spatial and
analysis of cells of different organisms that allowed for a temporal resolution – will allow extracting such a picture with
correlation to the phenomenological theory of evolution.[5–8] high confidence. Classical molecular dynamics simulations
Despite the fact that molecular biology addresses the provide efficient models and can be rigorously based on
atomistic resolution of biological structures and function and quantum mechanics (technically speaking, this is possible
may therefore be expected to be inherently quantum, it is through the Born-Oppenheimer approximation, which sepa-
typically not perceived as a field in which quantum mechanical rates electronic and nuclear motions, with the latter being then
phenomena play a dominant role. This is due to the fact that identified as the atomic motion in classical dynamics). Unfortu-
detailed molecular dynamics simulations rest almost exclusively nately, more detailed simulations of the quantum mechanical
on classical Newtonian mechanics. Moreover, the resulting equations are very difficult and only possible for a very small
qualitative understanding of the functioning of biomolecular number of atoms.
machinery, supported by extensive experimental results, often However, if we were to advance molecular simulations by
does not require a detailed atomistic structure and certainly not quantum computation fueled by the current hardware and
detailed atomistic dynamics. Instead, cartoonish representations algorithm developments in this field,[9–13] we may wonder to
of macromolecules are used in order to sketch a sequence of what extent biomolecular simulations[14] would actually benefit
events that they undergo. This knowledge alone is already from such developments and whether quantum computation
will become key in computational quantum molecular
[a] Dr. A. Baiardi, Prof. M. Reiher biology.[15–18] Brought to the point, the question is, whether the
Quantum for Life Center, Department of Chemistry and Applied Biosciences, emerging branches of quantum computation may eventually
ETH Zurich, Vladimir-Prelog-Weg 2, 8093 Zurich, Switzerland deliver a significant advance over traditional approaches.
E-mail: [email protected]
[email protected] Whereas an intensive search is ongoing for quantum effects
[b] Prof. M. Christandl on function in biology,[19–22] the most important quantum effects
Quantum for Life Center, Department of Mathematical Sciences, University are first and foremost rooted in the electronic structure of
of Copenhagen, Universitetsparken 5, 2100 Copenhagen, Denmark
E-mail: [email protected] biomolecules and, to a lesser degree, in their quantum nuclear
[**] A previous version of this manuscript has been deposited on a preprint motion (giving rise, for instance, to tunneling and kinetic
server (https://doi.org/10.48550/arXiv.2212.12220) isotope effects). The electronic structure of molecules is indeed
© 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH. This is key for the quantitative theoretical description and prediction
an open access article under the terms of the Creative Commons Attribution of chemical reactions in terms of reaction energies and
Non-Commercial NoDerivs License, which permits use and distribution in
any medium, provided the original work is properly cited, the use is non- activation barriers through the Born-Oppenheimer potential
commercial and no modifications or adaptations are made. energy surface (PES; see Figure 1). In other words, reaction

ChemBioChem 2023, 24, e202300120 (1 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

Figure 1. Graphical representation of the key steps involved in the simulation of a biological process. The solution of the electronic Schrödinger equation
yields the energy of a biomolecule (e. g., the light-harvesting complex II, top left part of the picture) for a given configuration of the nuclei. The change of the
electronic energy with the molecular structure (R1, R2, . . .) defines the potential energy surface (PES) and is represented as a blue surface for the electronic
ground state. A biomolecule is often a large and extremely flexible entity and its PES may display many minima, which correspond to stable structures (green
dots). Minima that are close on the PES correspond to conformers of the same molecule and are separated by a low-energy barrier that can be overcome with
thermal energy. Minima that are well separated across the PES and by higher energy barriers (red dot ) correspond to structures where the molecular
connectivity changes. The path joining them (red line) corresponds to a reaction. The ground-state PES can be approximated with a classical surrogate
GS
potential Eclassical (R) (bottom part of the figure) that is often referred to as force field. A force field expresses the PES, which results from the quantum
interaction between electrons and nuclei, in terms of effective interactions between atoms. One of its parts, Ebond(R), is associated with the bonding energy
between two atoms linked by a covalent bond. Two further terms are added describing the variation of the energy by modifying the angle between three
bonded atoms θj, Eangle(R), and the dihedral angle between four bonded atoms γm, Edihedral(R). Moreover, a force field will contain a term (EvdW(R)) that describes
the van-der-Waals interaction between pairs of non-bonded atoms that is, in turn, composed by the London attraction and the Pauli repulsion terms. Since
the PES changes with the state of the electronic system, electronic excitation by resonance with an external electromagnetic field (such as light, represented
as a yellow wave in the figure) switches to a new PES (we show only one of the excited electronic-state PESs in orange).

mechanisms of chemical processes in living cells can be system size. This curse of dimensionality creates a steep wall
analyzed and understood by computationally solving the that cannot be overcome with traditional computational
relevant quantum mechanical equations. Such calculations approaches (and therefore limits such attempts to small
provide insights into covalent bond formation in biomolecules systems of comparatively few atoms, which represent only
important to biochemical pathways, into proton and electron segments of biomolecules[23–25]).
transfer processes as well as into biological phenomena driven However, it is well known[26–30] that these types of
by visible and infrared light. interactions can be accurately modeled by classical surrogate
The fact that reversibility is key to molecular function in potentials for interacting atoms as new entities; typically only
biochemistry causes macromolecules to interact predominantly two fit parameters per pair potential are required as in Lennard-
through (many) non-covalent interactions. These weak inter- Jones potentials (note that no parameters whatsoever are
actions, such as van der Waals forces and hydrogen bonds, can required in explicit quantum chemical simulations of electronic
be easily broken under ambient conditions. They originate from structures). Such force-field approximations can be combined
the quantum correlations of electrons. However, these electron with algorithms that, being based on classical mechanics, do
correlation effects are very hard to describe in principle in a not suffer from the curse of dimensionality, but have a low
huge many-electron system such as a protein because the polynomial scaling with the number of atoms. The dramatic
computational effort for the solution of the underlying gain in efficiency is a result of the mapping of the electron
electronic Schrödinger equation scales exponentially with correlations onto pair interactions of atoms and thus mainly

ChemBioChem 2023, 24, e202300120 (2 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

due to the different algebraic structure of classical vs. quantum a redistribution of energy among kinetic and potential contribu-
mechanics. tions that eventually determine biomolecular function. Since
Such classical model potentials are the key ingredients of molecular biology rests on chemistry, it is also governed by the
biomolecular force fields in classical mechanics simulations of key theory of physics that quantitatively describes the electro-
bio-macromolecules, which approximate the Born-Oppen- magnetic interactions of electrons and atomic nuclei in and
heimer PES and facilitate the sampling of the high-dimensional between (bio)molecules, namely quantum electrodynamics.
configuration space through classical molecular dynamics or Starting from this basis, a quantum many-particle theory (see
Monte-Carlo methods in order to produce free energies from Section 3) and several computational solution methods (cf.
which the thermodynamic behavior of a biochemical system Section 4) have been devised.[31–34] As a result, energies and
can be deduced. Unfortunately, the underlying model poten- molecular structures can be predicted that allow us to describe
tials require sophisticated parametrizations, generally omit molecular behavior with breathtaking accuracy and depth.
quantum mechanical effects, lack transferability to other However, what can be accomplished in terms of actual
molecule classes, and therefore cannot describe well arbitrary computations depends on the size of the system, i. e., on the
chemical reactions and molecular interactions (if at all). By number of electrons and atomic nuclei. Very high accuracy can
contrast, none of these deficiencies is present in approaches only be achieved for very small molecules. Increasing molecular
that involve quantum mechanical effects, which are, however, size generally is accompanied by a loss in accuracy for feasibility
hampered by the highly unfavorable scaling problem with reasons, due to the high computational cost of exact quantum
system size. simulations. More and more approximations need to be invoked
In this work, we, therefore, discuss key computational target as the molecular size grows. Ab initio wave-function-based
problems in molecular biology from the perspective of methods aim at solving the many-body problem with approx-
quantum computation and elaborate on where and how to imate wave function parameterizations (such as truncated CI,[35]
achieve a quantum advantage in this field. The goal of this coupled cluster,[36] and perturbation theory[37]), which are
work is not only to deliver an overview of potential application routinely applicable to molecules with several dozens of atoms.
areas and refer to pioneering results already accomplished but Larger systems can be studied with methods that also
to give a specific roadmap on where to expect breakthroughs approximate the molecular Hamiltonian in terms of para-
that would not be accessible with traditional approaches. metrized models. A prominent example is approximate density
This paper is organized as follows: We first provide a tailored functional theory (DFT)[32,38] and related approaches (for in-
introduction to the foundations of computational physics stance, the random phase approximation[39]). The model param-
modelling applied to molecular biology in order to establish eters are often optimized to reproduce either highly accurate
some common ground in this highly interdisciplinary field for a quantum chemical or experimental reference data. Although
better understanding of the specific examples that follow. We many benchmark studies have set out to quantify the accuracy
begin with some principles of physical modeling of biomolecu- of different flavors of density functional theory, it remains hard
lar processes (Section 2); we then consider specific quantum to assess a priori the accuracy of a given density functional
equations (Section 3) and their traditional solution (Section 4), approximation for a specific molecule not included in the
before we dwell on the basics of quantum computation training data.[40]
(Section 5). Equipped with the key concepts and basic notation, For reasons to be discussed later, we may consider the size
we then turn to actual problems in molecular biology, where of a molecular system of a few hundred atomic nuclei to be
we first consider the theoretical description of chemical routinely treatable with approximate traditional quantum
reactions in biology (Section 6), followed by phenomena of chemical approaches. Obviously, such molecular sizes are far
quantum biology (Section 7) and molecular function amenable smaller than those of functional biological macromolecules
to classical molecular mechanics (Section 8). We finish (Sec- embedded in an aqueous phase.
tion 9) with remarks on the potential of data-driven methods To alleviate this drawback, embedding techniques have
for bioinformatics applications and eventually provide some been devised, which provide a quantum magnifying glass for a
general conclusions and outlook (Section 10). cluster of atomic nuclei and electrons considered to be decisive
for the quantum process within a bio-macromolecule (such as
the active site of an enzyme). The neglected part of the bio-
2. Physical modeling of biological processes macromolecule (and of the surrounding and penetrating water
molecules) is then treated as an environment. If part of the
The ways we describe, model, and understand processes in environment does not simply play a ’’spectator’’ role, but is
biology can be very diverse. If we consider molecular biology as reactive – such as (small) part of a protein environment (e.g., a
the core of biology, from which biological structures and solvent molecule or a few amino acid residues) – it becomes
functions, cells, and eventually the living world as a whole part of the quantum subsystem of the embedding scheme. The
emerge, achievements in chemistry and molecular physics can remainder of the environment is treated by a classical
be exploited to build a quantitative theory of the dance and mechanical model with its (possibly dynamic) partial charges
interactions of biomolecules. and geometric constraints affecting the quantum region
In this ansatz, the dynamical behavior of key entities is through an additional electrostatic potential. The determination
described by a mechanical equation of motion, which allows for of this quantum region can be automated (see, e. g., Refs. [41]

ChemBioChem 2023, 24, e202300120 (3 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

and [42]). Numerous approaches have been devised for the 3. Theoretical foundations: Quantum electronic
embedding of a small active quantum part into an environment and vibrational structure
(see, e. g., Refs. [24, 43–53]). We note, however, that large-
amplitude motions and long-range effects may be difficult to
describe in such settings. Molecular motion is governed by the laws of quantum
In terms of chemical diversity, biomolecules are mostly mechanics. Defining these laws and the equations they lead to
organic molecules, often with rather regular bonding patterns represent the first step toward understanding the potential of
(such as proteins, polysugars, nucleic acids, and other biopol- quantum computing for molecular simulations. Quantum
ymers). It is for this reason that electronic quantum correlations mechanics describes a molecule through the molecular wave
turn out to be rather regular in the sense that they can be function Yðr; R; tÞ, which is a complex-valued function that
mapped to simple classical models. These classical models are depends on the positions of all its elementary particles, i. e. the
much easier to handle in actual computations, and therefore, electrons (r) and the nuclei (R), and on time t. The Born rule
much larger molecular sizes can be accessed (such as large states that the squared norm of the wave function, jYðr; R; tÞj2 ,
proteins and protein complexes). defines the probability density of finding the molecule in a
In other words, the underlying Coulomb interaction of given configuration ðr; RÞ at a given time t. The second key
electrons and nuclei within and between biomolecules can be quantity is the molecular Hamiltonian H ^ ðr; R; tÞ, which deter-
mapped onto specific forms of potential energy functions, now mines the time evolution of the wave function through the
defined between new dynamical entities: atoms, united atoms, time-dependent Schrödinger equation,
or even larger coarse-grained aggregates. Surprisingly trans-
ferable potential energy functions are obtained that capture the ^ ðtÞYðr; R; tÞ ¼ i�
H h@t Yðr; R; tÞ : (1)
fundamental interactions in a molecular mechanics framework,
which is now rooted in classical Newtonian mechanics rather As shown in Figure 2, the Hamiltonian describing a mole-
than quantum mechanics. Prototypical analytical forms for the cule contains contributions associated with the kinetic energy
distance-dependent potential energies are parametrized elec- of the atomic nuclei and of the electrons as well as the
trostatic and Lennard-Jones model potentials. Only if chemical interaction between these particles. The interactions are very
bonds are formed or broken, such descriptions become more well described by pairwise electrostatic Coulomb potential
involved and require so-called reactive force fields[54] that are, energies, although they neglect, for instance, magnetic inter-
however, highly specialized and tailored towards a given class actions. Since these interactions do not depend on time, we
of molecules. Moreover, their accuracy is often difficult to assess can rigorously eliminate the t dependence of the Hamiltonian,
– as is the case for any methods based on classical surrogate ^ ðtÞ ! H.
i. e., H ^ Now, any solution to Eq. (1) can be expressed in
potentials. Due to the lack of generality and possibility of terms of specific wave functions, often referred to as stationary
reliably controlling the error, only explicit quantum mechanical states, whose probability amplitudes do not change with time.
calculations can deliver reliable insights into any biochemical These wave functions, Yn ðr; RÞ, are the solutions to the time-
process that is accompanied by breaking and forming independent Schrödinger equation,
bonds.[55–57]
All of these approaches refer to the same basic concept, ^ n ðr; RÞ ¼ En Yn ðr; RÞ :
HY (2)
namely that of a potential energy hypersurface, i. e., the PES,
which assigns an energy to any atomic configuration of atomic i. e. they are the eigenvectors of the Hamiltonian H ^ with
nuclei/atoms in a molecule, and hence, allows one to judge eigenvalues En, their energy, assigned to them.
structural changes in terms of a change in potential energy. The Molecular machinery in biology is viewed as intricate
(quantum) energies that constitute the PES are the electronic systems of numerous molecules, whose actions are governed
energies defined in the Born-Oppenheimer approximation mostly by reversible interactions of the biomolecules contained
which assigns energy to a fixed scaffold of atomic nuclei, in in the systems. Examples are hydrated protein complexes
which the electrons of the molecule move. Depending on the embedded in a membrane (such as ATPase) or ribosomes. It is
well depths, the number of wells, and the barrier heights because of the weakness of these reversible interactions that
exhibited by this hypersurface, different physical modeling quantum mechanical modeling can efficiently address the
approaches can be activated (see Figure 1). quantum states of the individual molecular players (and their
The key take-home message is that we can describe the interactions) rather than describing all of them as a whole
quantum mechanical behavior of electrons and nuclei in system; i. e., separated biomolecules and their interactions are
biomolecular systems in principle, but often not in practice, as studied rather than an organelle or a cell as a whole. While this
such calculation would be too computer-time demanding. In separation into components makes the solution of the
such cases, one resorts to classical models and accepts a Schrödinger equation then feasible (at least for not too large
reduced accuracy and a loss in transferability. The promise of molecular structures), it also allows one to piece together and
quantum computations is that a quantum description can be eventually understand biomolecular machinery in terms of
extended to larger atomistic structures and to higher accuracy individual energies assigned to the states of the molecular
(compared to approximate traditional quantum chemical players. For this reason, Eq. (2) is the main cornerstone of the
approaches). simulation of biochemical processes.

ChemBioChem 2023, 24, e202300120 (4 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

Figure 2. Graphical representation of the key equations for quantum mechanical molecular calculations. Electronic properties are described, for a given
configuration of the nuclei R, by the electronic Hamiltonian, H ^ R defined in the upper left part of the figure. H^ R is obtained from the molecular Hamiltonian
ele ele
by neglecting the nuclear kinetic energy, i. e. by adopting the fixed-nuclei approximation. It contains, therefore, the kinetic energy of the electrons T^e (r), the
electron-electron Coulomb interaction V^ee (r), the nuclear-electronic Coulomb attraction V^neR
(r), and the nuclear-nuclear Coulomb repulsion V^nn R ^R
. Hele depends on
the electronic coordinates r and, parametrically, on the nuclear ones R, because it does not contain any derivative operator with respect to R. The
corresponding electronic ground and excited states wave functions (YRele;GS (r) and YRele;ES (r)), together with the corresponding energies (Equantum;GS
R
(r) and
R
Equantum;ES (r)) will display the same parametric dependence. The variation of the electronic energy of a given electronic state with the nuclear coordinates
defines the PES (represented as a green curve for the electronic ground state and as a red curve for the lowest-energy excited state). The PES is usually only
calculated point-wise (black crosses) and approximated with fitting algorithms. We approximate the green PES around the left local minimum by a parabolic
function (the corresponding curve is displayed with a dashed line), which defines the harmonic approximation. The vibrational Schrödinger equation for the
jÞ
electronic ground state is reported in the lower part of the figure. Its solution yields the vibrational wave functions Yðnuc;GS . Since the PES changes with the
electronic state, a different set of vibrational wave functions will be obtained for the excited state.

R
Already for a molecule with a few dozen atoms, the wave the Schrödinger equation for a Hamiltonian containing Equantum
function depends on hundreds of variables, and Eqs. (1) and (2) as interaction term (see Figure (2)). The overall molecular wave
become extremely hard to solve on classical computers. The function is then given by Ynuc ðRÞ � YRele ðrÞ.
Schrödinger equation can be simplified by invoking the Born– Whether one needs to solve either the time-independent or
Oppenheimer approximation: since the mass of the nuclei is at the time-dependent Schrödinger equation, either only for the
least three orders of magnitude larger than that of the electrons or also for the nuclei, depends on the biochemical
electrons, the nuclear dynamics are much slower than the process at hand. In all cases, the wave function is the solution
electronic ones. It is, therefore, reasonable to calculate the wave of a differential equation in which all variables are coupled by
function in two steps. First, the electronic wave function is the potential energy operator. This coupling correlates the
calculated by neglecting the kinetic energy contribution of the motion of the quantum particles and renders the solution of
nuclei. This yields the so-called fixed-nuclei Hamiltonian H^ R (see the differential equation a daunting computational task for
ele
Figure 2), where the nuclear coordinates R do not enter as large molecules.
variables, but rather as fixed parameters. Therefore, the Why is the solution of many-body quantum problems such
R
eigenvalues Equantum and eigenfunctions YRele ðrÞ depend para- a major computational hurdle? We consider the electronic
metrically on the nuclear coordinates R, which we therefore problem, although our conclusions will apply to the nuclear
indicate as a superscript. They describe the allowed states and one as well. In order to represent it on hardware, either classical
the corresponding energy of the electrons for a given or quantum, the many-electron wave function must be
configuration of the nuclei R. The nuclear part of the molecular discretized. This discretization is realized by expressing the
wave function is then calculated by assuming that the change wave function in terms of a finite set of one-electron functions
in the nuclear coordinates is slow enough to ensure that the ffk ðr; s ÞgNk¼1
orb
, the so-called molecular orbitals. Specifically, the
electrons are described by the same eigenstate (e. g. the ground wave function is expressed as a linear combination of
state) YRele ðrÞ for all R values. This approximation can be seen as antisymmetrized products of molecular orbitals as sketched in
a result of the adiabatic theorem of quantum mechanics.[58] The Figure 3. An antisymmetrized product is often referred to as
R
eigenvalue Equantum associated with the eigenfunction describing Slater determinant or as electronic configuration. Methods that
the electrons defines the effective interaction potential be- solve the Schrödinger equation within the resulting wave
tween the electrons and the nuclei. The nuclear wave function function ansatz without any restriction of the space of
Ynuc ðRÞ describing this motion is, therefore, obtained by solving configurations are, in chemistry, referred to as full configuration

ChemBioChem 2023, 24, e202300120 (5 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

Figure 3. Graphical representation of how the full-CI wave function is constructed starting from a given set of molecular orbitals for the butadiene molecule.
The full-CI wave function Ψele(r1, . . ., rNele ) is expressed as a linear combination of antisymmetrized products of the molecular orbitals φi(r) (a graphical
representation of the set of molecular orbitals of ethylene is given in the right upper part in blue and red, indicating the different signs of these functions).
From each molecular orbital, two spin orbitals, one for spin-up and one for spin-down electrons, are constructed. The antisymmetrization operator A is
applied to each configuration to ensure that each of these basis functions is antisymmetric upon exchange of electrons – as prescribed by the Fermionic
nature of the electrons. The coefficients entering the linear combination are stored in the tensor Ci1 ;:::;iN , which has one index per electron. By virtue of the
ele

Pauli exclusion principle, a spin orbital can be occupied by at most one electron. Therefore, as shown in the upper part of the figure, the overall number of
terms included in the linear combination is equal to the number of subsets of Nele elements of the molecular orbital set, which contains Norb elements. This
binomial scaling results in the curse of dimensionality.

interaction (full-CI), because the wave function is expressed as a Encoding an extremely complex high-dimensional wave
linear superposition of all electronic configurations. In physics, function in terms of much simpler entities, the orbitals, comes
this is known as exact diagonalization. The molecular orbitals with a price. Let us consider a molecule with Nele electrons
are themselves discretized as a linear combination of a finite set described by Norb molecular orbitals. Let us also assume that
of orbitals – usually referred to as atomic orbitals to distinguish Norb is proportional to Nele. This is a reasonable assumption
them from the molecular ones. The atomic orbitals are basis because, as highlighted above, Norb is proportional to the
functions that are pre-optimized to represent the orbitals of a number of atoms of the molecule that is, in turn, proportional
given atom (and its polarization) accurately. They are expressed, to Nele. Since one can choose any one of the Norb molecular
in practice, as a linear combination of Gaussian basis functions. orbitals to describe each electron, the number of configurations
This facilitates the calculation of integrals that define the that can be constructed grows combinatorially with Nele. This
representation of the molecular Hamiltonian in the atomic implies that the computational representation of the wave
orbital basis. function requires storing a number of coefficients that grows
For a molecular system, the union of the atomic orbitals of exponentially with the number of electrons. The resulting
all atoms defines the atomic basis in which the molecular unfavorable scaling is often referred to as the curse of
orbitals are expanded. The expansion coefficients are called dimensionality and renders full-CI impractical on classical
molecular orbital coefficients, which can then be optimized hardware already for molecular systems with more than three
based on the variational principle. When the optimization is to five atoms.
only carried out for a single Slater determinant, this yields the The curse of dimensionality plagues any full-CI-based
Hartree-Fock (HF) method. method aiming at solving quantum many-body problems in
In the limit of a dense, complete atomic basis set – a limit terms of one-particle functions. This hurdle is, therefore, not
called the complete basis set limit – the many-electron wave limited to time-independent electronic calculations. The same
function can be encoded exactly as a linear superposition of challenges will be faced for the solution of the time-dependent
Slater determinants. The complete basis set limit cannot be Schrödinger equation, both for the nuclear and for the
reached in numerical simulations that must necessarily rely on electronic case. The form of the differential equation changes –
finite, incomplete basis sets. Extrapolation techniques have the time-independent Schrödinger equation is an eigenvalue
been designed to tame this problem. These methods assume equation, while the time-dependent one is a partial differential
that the electronic energy Eele depends on the basis set size n equation that involves both the spatial and the time variables.
according to a given, fixed power law. One can then solve the Still, both equations are defined in terms of the Hamiltonian
electronic Schrödinger equation for increasingly larger basis operator that couples the variables on which the wave function
sets, determine the power law coefficients by least-squares depends. Full-CI is, therefore, needed to describe the resulting
fitting, and estimate the complete basis set electronic energy quantum correlations completely, leading once more to the
Eele from the limiting value obtained for n ! þ∞. These emergence of the curse of dimensionality.
methods are, however, inherently heuristic because the form of Research in quantum chemistry has mostly focused on
the power law must be fixed a priori. taming the curse of dimensionality with algorithms that trade
the full-CI accuracy and generality for computational efficiency.

ChemBioChem 2023, 24, e202300120 (6 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

Specifically, they rely on wave function approximations which electrons coalesce. It can be proven that, at the coalescence of
scale polynomially in the number of expansion parameters but two electrons, the many-body wave function displays a cusp
are tailored to a specific regime of electron correlation. In (the so-called Kato cusp), i. e. its derivative is not continuous.
quantum chemistry, the term electron correlation denotes an Attempting to fulfill the Kato cusp condition with products of
ensemble of effects that are not accounted for by the Hartree– cuspless Gaussian basis functions yields highly non-compact
Fock ansatz. This notion corresponds well with notions of full-CI wave functions, with many configurations with small but
correlation and entanglement in the context of quantum non-negligible coefficients – the signature of dynamical
information theory, the number of necessary configurations correlations. Accordingly, the basis set expansion of the wave
corresponding to the Slater rank[59] – the Fermionic analog of function is often supplemented by factors that depend
the Schmidt rank quantifying quantum entanglement.[60] More explicitly on the interelectronic distance[69]
detailed information about the weight components can be cast
as eigenvalues of the one-particle reduced density matrix with
naturally associated entropic measures and relation to Pauli’s
principle.[61–64] 4. Computational foundations: Traditional
Consider the idealized case in which the Coulomb repulsion approaches
between the electrons is vanishing and, therefore, the electrons
are not coupled. In this case, the full-CI wave function coincides The curse of dimensionality limits practical applications of full-
with the Hartree-Fock determinant. Therefore, electron correla- CI to molecules with approximately 20 electrons,[65,70] a size that
tion effects are those that impede encoding the wave function characterizes very small molecules, far smaller than any
as a single determinant and require it to be expressed as a biomolecule. The impossibility of routinely applying full-CI to
linear superposition of configurations. large molecules has sparked the development of approximate
Qualitatively speaking, the larger the number of config- methods not suffering from the curse of dimensionality. The
urations with non-negligible expansion coefficients, the stron- key idea behind many of these approximations is to identify
ger the correlation effects will be. Paradigmatic examples of wave function parametrizations that can approximate full-CI
strong correlation effects in molecules are transition metal while being defined by a number of parameters that scales
complexes (as they occur in active sites of metalloenzymes and polynomially with the system size. These approximate para-
FeS clusters) and bond-breaking processes.[65–67] metrizations have been tailored to two idealized situations:
In the former case, the manifold of d-type atomic orbitals of weakly correlated and strongly correlated electronic structures
transition-metal active sites, and transition metal complexes in of molecules. Unfortunately, strong and weak correlation effects
general, often contains near-degenerate orbitals that are often coexist in a molecular system and, therefore, these
partially occupied. The corresponding configurations will, there- methods must be combined in accurate molecular simulations.
fore, have a similar weight in the full-CI wave function. Similarly, Above, we defined weakly correlated wave functions as
bond-breaking processes yield electronic configurations where those in which the Hartree–Fock configuration has a predom-
unpaired electrons are localized on different parts of the inant weight (given as the square of its expansion coefficient in
molecule – a situation that cannot be captured by a single the expansion of configurations). Since the wave function is
configuration in a singlet state because spin pairing will always normalized to a fixed value (usually one), all other determinants
produce two degenerate configurations with alpha and beta will have a comparatively small weight. Moreover, it is
spin orbitals exchanged on the dissociating parts. reasonable to expect that the weight of the other configu-
Moreover, weak electron correlation effects – referred to as rations will decrease as their departure from the Hartree-Fock
dynamical correlations in quantum chemistry – must also be determinant increases. This departure can be measured in terms
taken into account. Although they do not alter the qualitative of so-called excitations (which are orbital substitutions see
structure of the full-CI wave function, they yield a non- Figure 4): a determinant will be defined as a Nsub-fold excited
negligible contribution to the electronic energy because of the determinant if it differs from the Hartree-Fock reference one by
sheer number of configurations with very small weights. at most Nsub orbitals. We note that the term excitation is
Dynamical correlations comprise two classes, short- and misleading, despite having become a standard notion, as it
long-range correlations. From a physical perspective, long- implies that an orbital substitution corresponds to an electroni-
range correlations are associated with the London dispersion cally excited state, which is only true for non-interacting
forces, which, qualitatively speaking, describe the induced electrons and obvious when considering that the ‘excitations’
dipolar interaction between two molecular fragments that are still elements of a many-electron determinantal basis set
result from instantaneous polarization of the corresponding into which the ground-state wave function is expanded in the
electronic density. Although they require, in principle, a proper configuration interaction ansatz. Based on this idea of orbital
quantum description of the electron-electron interaction, dis- substitutions, the full-CI wave function can be systematically
persion interactions can be well-approximated with effective approximated by including only up to Nsub-tuple substitutions
pairwise classical potentials.[68] in the wave function expansion. This leads to so-called
Short-range correlations are, instead, associated with the truncated CI approaches, whose computational complexity
short-range electron-electron interaction, i. e., in the limit of a scales polynomially with the number of orbitals, with an
vanishing interelectronic distance in which the positions of two exponent proportional to Nsub. We note that the emerging

ChemBioChem 2023, 24, e202300120 (7 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

Figure 4. Graphical representation of configuration interaction-based wave function approximations (illustrated for the electronic wave function of butadiene).
Configuration Interaction methods encode the many-body wave function (upper part of the figure) in terms of one-particle functions, known as orbitals in
electronic structure theory (bottom left part of the figure). Molecular orbitals may be optimized with Hartree-Fock or Density Functional Theory and are
expressed as a linear combination of a Norb-dimensional basis set. Out of the Norb molecular orbitals obtained from such a calculation, the Nele/2 ones that yield
the lowest-energy determinant are called occupied and the other ones are denoted virtual. The occupied orbitals define the reference determinant (a
determinant is often referred to as configuration). The configurations entering the full-CI wave function can be classified in terms of the number of orbitals by
which they differ compared to the reference determinant. Configurations differing by a single spinorbital are referred to as ‘single excitations’ (represented in
gray), while the ones differing by two spin orbitals are referred to as ‘double excitations’ (represented in yellow). Truncated CI approximates the full-CI wave
function by including only all those configurations up to Nsub-fold excitations. For Nsub = 1, one obtains the configuration interaction singles (CIS)
approximation, whereas for Nsub = 2 the method is referred to as configuration interaction with singles and doubles excitations (CISD). If Nsub coincides with
the overall number of electrons of the molecule, one retrieves full-CI. Note that the general notation for the expansion coefficients Ci1 ;:::;iN in Figure 3 is here
ele

replaced to follow the partitioning according to excitation classes as Cia ; Cijab , …

hierarchy of approximations is an ad hoc choice that does not classes: selected configuration interaction methods and algo-
guarantee that all truly important configurations are actually rithms based on tensor network states.
included in the truncated expansion of configurations. The key Selected configuration interaction methods rely on the
limitation of truncated CI (by contrast to full-CI), however, is assumption that, although not being dominated by a single
that it is not size extensive, i. e. the corresponding energy does configuration, the full-CI expansion remains highly sparse so
not scale linearly with the number of electrons,[71] which makes that the number of negligible expansion coefficients Ci1 ;...;iN ele

it impossible to reliably compare the energy of molecules with (see Figure 3) is in overwhelming majority. Full-CI Quantum
different numbers of electrons. Monte Carlo (FCIQMC)[74] exploits this idea by solving the
The established canonical solution to this problem is the Schrödinger equation via a stochastic Monte Carlo-based
coupled cluster (CC) method.[72] CC relies on an exponential realization of the imaginary-time evolution algorithm. Deter-
wave function ansatz that contains a number of parameters ministic variants of FCIQMC are also available,[75–79] which
that scales polynomially with system size. Specifically, CC with originate from the “Configuration Interaction using a Perturba-
singles, double, and perturbative triple excitations, abbreviated tive Selection made Iteratively” (CIPSI) method introduced by
CCSD(T), has become the de facto reference method for Malrieu[80] that identifies the non-negligible configurations
calculating weakly correlated electronic wave functions in order based on second-order perturbation theory. In practice,
to reach chemical accuracy of about 1 kcal/mol. selected CI methods can target up to 100 electrons, which is a
Approaches such as truncated CI or CC – usually referred to huge leap forward for chemical applications, but still represents
as single-reference methods, because they rely on a single a severe limitation for generally large biomolecules. Surely,
configuration, the Hartree–Fock one, as a reference – can other approximate quantum approaches that do not rely on
become inaccurate as soon as the full-CI wave function is not expansions of configurations – most importantly contemporary
dominated by a single configuration. For such multi-reference methods of Density Functional Theory (DFT; see below) – can
cases, modern full-CI-type methods[65,73] have made it possible target systems that are two orders of magnitude larger, but
to push the limits of full-CI to systems with approximately 100 they also come with an uncontrollable error for a specific
electrons. These methods can be qualitatively divided into two biomolecule under consideration.[40]

ChemBioChem 2023, 24, e202300120 (8 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

An alternative route is offered by tensor network states improving the accuracy of a DFT calculation, which is the major
(TNSs).[81] In essence, a tensor network approximates the tensor drawback of contemporary DFT.
C ¼ Ci1 ;...;iN of all full-CI expansion coefficients (cf. Figure 3) with
ele
We pause at this point for the conceptual observation that
tensor compression algorithms. This tensor compression re- the ground state energy can be written in variational terms as
duces the number of wave function parameters but is inevitably
associated with a loss of accuracy. However, TNSs can be Eg ¼ min hyjHjyi (3)
designed to yield an optimal balance between compression
degree and accuracy. In many numerical studies it turned out and that methods that optimize over an ansatz (variational
that a specific class of TNSs, so-called matrix product states, methods) such as Hartree-Fock, full-CI, DFT, restrict the
reliably approximate the electronic Schrödinger equation,[65,81–90] minimization and always lead only to an upper bound on the
whose parameters can be very efficiently optimized with an true ground state energy. Whereas efforts can be made to
iterative algorithm known as density matrix renormalization assess the closeness to the true ground state energy (e. g.
group (DMRG) approach.[91,92] Although we have focused so far extrapolation to an infinite basis set in the case of full-CI), there
on the key problem of molecular quantum simulations, namely can ultimately not be any rigorous closeness estimate of such
the solution of the time-independent electronic Schrödinger methods (as the infinite basis set limit cannot be reached in
equation (cf. Figure 2), we note that the approaches discussed practice) or the ansatz within a given basis set only samples a
so far can be generalized to the nuclear problem as well (see, subset of the states (e. g. Hartree-Fock, CC, …).
for example, Ref. [93] for the introduction of vibrational DMRG) It was already noted in 1960 by Coulson[100] that rigorous
and to the time domain to solve time-dependent Schrödinger lower bounds could be obtained by the observation that
equations.[94,95] molecular Hamiltonians only involve pairwise interactions
P
Even for complex molecules with several dozen atoms, H ¼ hðij2Þ and that, therefore, the ground state energy can be
ij
electronic correlation is rarely strong for more than 100
cast as an optimization over the two-particle reduced density
electrons. This has inspired the development of so-called active-
matrix (2-RDM) 1ð2Þ
space algorithms: a full-CI-based scheme is applied only on the
strongly correlated orbitals, while the remaining weak correla- !
tion effects are captured a posteriori. Translating such an N �
Eg ¼ min Tr hð2Þ 1ð2Þ : (4)
apparently simple idea into an algorithm is a hard task. In fact, 2
designing algorithms to include accurately weak correlation
effects on top of a DMRG or FCIQMC simulation remains an
open challenge of computational chemistry. One should not The function to be optimized is linear and the set of 2-RDMs
forget that the second hurdle of active space-based methods is is convex and only of polynomial dimension in the basis set,
the need of identifying the strongly-correlated electrons a priori, sparking hope for a solution to the problem. The earlier
without knowing the exact full-CI wave function. In this respect, discussed difficulty of the ground state energy problem is now
various strategies have been proposed to efficiently identify hidden in the complexity of describing this convex set. The
strong correlation effects based either on single-reference concrete question is: Given a matrix 1ð2Þ , is it the reduced
calculations or on partially converged full CI-type density matrix of an N-particle electronic wave function? This
simulations.[63,96–99] problem is known as the N-representability problem of
Finally, we emphasize that DFT represents a route to solve quantum chemistry and solving it as Coulson’s challenge.[100–102]
the electronic structure problem that has become the work- Whereas the exact characterization of the set has been
horse of computational molecular quantum mechanics. The key shown to be Quantum Merlin Arthur (QMA)-hard (QMA is the
difference between a superposition of configurations and DFT quantum analog of the famous class of computational problems
is that DFT calculations in practice rely on the Kohn-Sham known as nondeterministic polynomial time (NP)) and thus
variant, where the density 1ðrÞ is obtained from a single unlikely to be solved even on a quantum computer,[103] we note
determinant wave function in such a way that it can be that any outer approximation to the set will result in a lower
calculated as a sum of absolute squares of the orbitals entering bound on the energy, in stark contrast to the situation
the determinant. The key insight here is that one can obtain the discussed above for the variational methods.
exact ground-state electron density of an interacting quantum Concretely, such approximations can be given by semi-
system from this determinant, which describes exactly a non- definite constraints (rendering the optimization problem a
interacting surrogate system of fermions. The orbitals are semi-definite program) on the set and a number of nontrivial
obtained by solving a Hartree-Fock-like set of equations that is constraints have been found and, in general, significant efforts
defined in terms of the so-called exact density functional have been devoted to this problem.[104–106] For chemical
E½1ðrÞ�. While this procedure is exact, the exact form of E½1� is, problems, unfortunately, the rigorous lower bounds obtained
unfortunately, not known. Practical DFT calculations rely on with this method have not been as close from below to the
heuristic forms of E½1� that are either data-driven or inspired by believed true ground state energy as the variational methods
the form of the functional obtained for simple model systems. from above. The success of such methods has therefore been
Estimating the accuracy of approximate functionals is often limited in practice.
impossible,[40] which prevents quantifying and systematically

ChemBioChem 2023, 24, e202300120 (9 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

Many of the methods outlined so far target the lowest- ents in a quantum computer are most naturally cast in terms of
energy solution of the electronic Schrödinger equation (see quantum bits (qubits), quantum mechanical two-level systems.
Figure 2) which corresponds to the electronic ground state. The Qubits in a quantum computer are combined with the tensor
ground-state energy is the key quantity for reactions that product operation. The idea to build a quantum system to
happen on the electronic ground-state PES (see Figure 1). simulate another quantum system – that quantum simulates
However, some biological processes are driven by the absorp- quantum – is as self-evident and simple as it is revolutionary.
tion of electromagnetic radiation. In this case, the reaction The technical reason that hinders classical simulators at the
involves the surface of an electronically excited state, as shown possibility of simulating quantum systems is the different
in Figure 1, which corresponds to higher-energy solutions of composition rules of its elementary components. In fact, the
Eq. (2). These solutions are not only important to describe a classical composition rule is weaker than the quantum mechan-
biomolecular system after light excitation, but they also offer ical. This means, in practice, that only a subset of the possible
information that can be related to spectroscopic techniques quantum mechanical states of N qubits can be represented by
and, hence, to experimental spectroscopies that identify specific N of their classical counterparts. This is precisely remedied by
states of biomolecular machinery. Many of the methods building a simulator directly from quantum mechanical constit-
reviewed above can be straightforwardly generalized to target uents, since those evidently adhere to the same quantum
excited states – although these generalizations are often harder mechanical composition rule and, therefore, offer a computa-
than their ground-state counterparts. tional platform perfectly suited to the task of simulating a
Molecules can be subjected to spectroscopic techniques quantum system.
through their response to external electromagnetic perturba- By simulating a quantum system we mean, just as before,
tions (i. e., light). The spectroscopic response of a molecule does simulating relevant properties of quantum systems – the most
not depend only on its energy, but also on its properties, such relevant being, in chemistry and biology – molecular properties.
as the electric and magnetic dipole moment. Response proper- Primary examples of relevant quantities to be simulated are, for
ties can be calculated based on the methods listed above with static phenomena, ground- and excited-state energies and, for
time-dependent perturbation theory.[107] non-equilibrium processes, time-dependent correlation func-
Identifying the method providing the best compromise tions. Simulating quantum systems is not the unique applica-
between cost and accuracy for time-dependent simulations is tion of quantum computing. Other computational problems
an especially hard task. When driven out of its ground state, the that are relevant for physics, chemistry or biology and that are
nature of the wave function and, therefore, the correlation hard to solve on a classical computer may be solved more
between its particles may change, even dramatically. It is known efficiently on quantum computers. A candidate in the biological
that the time evolution of a wave function following an external context is the protein folding problem, which is an optimization
perturbation (usually known as quench in quantum physics, problem, for which heuristic quantum optimization algorithms
which corresponds most often to an electromagnetic perturba- are being investigated.[109,110]
tion in chemistry) may lead to a continuous growth of its It is important to emphasize that, although quantum
entanglement. This makes it difficult to accurately represent the simulates quantum is such a self-evident slogan, the existing
time-evolving wave function as a TNS over the whole quantum simulation algorithms are highly non-trivial and
propagation. For example, the entanglement growth makes the problem-dependent. Their future development will therefore
MPS representation of the wave function increasingly inaccu- require ingenuity and insight both from the general principles
rate with increasing time. How this loss in accuracy affects the of quantum algorithmic design (mathematics, physics, and
simulation accuracy depends on the simulation target at hand. computer science) and from the application side (physics,
chemistry, biology).
Quantum simulators naturally fall into two categories,
traditionally called analog and digital quantum simulators.
5. Computational foundations: The promise of Analog quantum simulators are more akin to purpose-built
quantum simulation machines that through their build-up mimic the interactions
and, therefore, the behavior of the system to be simulated.
It was Feynman[108] who noticed that using a classical system, When equipped with knobs to modify the behavior of the
e. g. a traditional computer, to simulate a quantum system will simulator they can have aspects of universality.[111,112] Yet, they
in general result in an exponential overhead in the time are meant as purpose-specific computing machines and, there-
required for the simulation due to the curse of dimensionality, fore, are not meant to be universal computing machines.
but that building a simulator (or computer) directly from Mimicking the interactions of a quantum system with an analog
quantum mechanical constituents (henceforth called a quan- quantum simulation is often hard due to the fact that the
tum simulator or, if very versatile, quantum computer) could many-body interaction terms of a given Hamiltonian (for
circumvent this curse and therefore achieve an efficient instance, the electronic Hamiltonian of Figure 2) may not be
simulation. With efficient simulation we mean a simulation that directly encoded on the hardware at hand. This limit can be
requires time and space resources that scale proportionally (or overcome with so-called gadgets that map, via perturbation
at most polynomially) with the size of the biomolecule to be theory, the desired interactions of the system to be simulated
simulated. Analogous to the bits of a computer, the constitu- to effective interactions built from those available in the

ChemBioChem 2023, 24, e202300120 (10 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

hardware. The price to be paid is a slight increase, usually a contexts is that the (discretized) interaction is drastically more
multiplicative constant, of the hardware size needed to simulate irregular in the latter case than in the former one. Tunability is,
a given quantum system. therefore, a key requirement of an analog quantum simulator
The amazing experimental progress, especially in the area for quantum chemical and quantum biological simulations.
of optical lattices, where many atoms are trapped and interact In a nutshell, analog quantum simulators are single-purpose
in a lattice of standing light waves, has delivered remarkable machines with an easily scalable size designed to mimic the
simulation results of condensed matter systems surpassing the behavior of interesting quantum systems. Affected by decoher-
practical simulation capability of traditional computers.[113,114] ence on the individual-component level and without complete
Yet, since the constituents of analog quantum simulators also error-correcting and fault-tolerant mechanisms in place, how-
experience decoherence – a quantum form of errors – which ever, few rigorous computational guarantees are available.
cannot be suppressed arbitrarily (this is true at present, and Digital quantum simulators are, by contrast, based on the
may remain so even in the far future), formal guaranteed gate-based model of quantum computation in direct analogy to
statements about the overall accuracy of the achieved the gate-based model of standard (classical) computation (see
simulation are difficult if not impossible. Control in such a the upper part of Figure 5). Here, a quantum algorithm is cast
system is continuously being improved, yet, in general, as a sequence of quantum gates, i. e., operations acting on
experimental analog quantum simulators are characterized by quantum bits (qubits). Qubits are the analog of the bits in a
limited control and accuracy, but have a relatively easily standard computer, but can, in addition to the states 0 and 1,
scalable number of constituents. The limited control constitutes also assume a quantum superposition of both, in the same way
a major challenge when simulating (bio)chemical processes that an arrow cannot only point up or down but also in any
with digital simulators. The key difference between Hamilto- other direction. In a physics context, qubits are also known as
1
nians relevant in quantum physical problems and those two-level systems and can often be viewed as a spin-2 system.
appearing in quantum chemical and quantum biological Importantly, the composition rule of quantum particles, and

Figure 5. Graphical representation of the steps involved in the simulation of a molecular process with digital and analog quantum simulation. For a given
biochemical target (such as the LHC-II complex in the upper left part of the figure), one first chooses the Hamiltonian H ^ mol that describes the process at hand.
For LHC-II, this could be the excitonic Hamiltonian describing the excitation energy transfer process. Assume we aim at solving the time-dependent
Schrödinger equation (box in the left part of the figure) for a given initial wave function jY ð0Þi. In digital quantum simulations (upper part of the figure), one
first maps the target quantum system to a set of Nq qubits (in the figure Nq = 5), and maps H ^ mol onto a corresponding qubit Hamiltonian. A sequence of gates
is then defined to solve the time-dependent Schrödinger equation. In the figure, we selected the specific case of the Trotter algorithm for simulating quantum
dynamics (see Eq. (5)). In analog quantum simulations (lower part of the figure), the quantum system is first mapped onto a set of interacting quantum
entities (such as, for instance, atoms trapped in arrays of optical tweezers). The interaction between these entities is then tuned (for instance, with a laser, as
represented in the figure) such that the Hamiltonian describing these interactions (H ^ sim ) mimics the original Hamiltonian H^ mol . The quantum dynamics is then
simulated directly on the resulting simulator, without the need of designing any algorithm for solving the time-dependent Schrödinger equation. The result of
a simulation is a time-dependent property (O)(t), as represented in the right part of the figure. The analog quantum simulation does not admit rigorous
approximation bounds yet aims to give an accurate qualitatively reliable simulation for a long propagation time. The digital simulation, by contrast, will track
the property with high accuracy, yet only for a shorter period of time due to the stringent hardware requirements of fault-tolerant quantum computation.

ChemBioChem 2023, 24, e202300120 (11 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

hence also of quantum bits, posits that the state of n qubits mapping is often done with the Jordan-Wigner[121] or the Bravyi-
requires the description with 2n probability amplitudes (i. e., Kitaev transform,[122,123] but further algorithms-tailored methods
expansion coefficients; see Figure 3) describing the precise are expected to be developed in the future. Common to all
superposition of the strings of n bits (thereby making up the ‘ simulation algorithm designs is the drive to the most favorable
state’ or ‘ wave function’ of the n qubits). Simulating or even scaling of simulation time with the size of the simulated system
just storing the amplitudes on a standard computer, therefore, in order to make actual quantum computations feasible.
requires more than 2n bits and therefore an exponential An important class of algorithms for NISQ hardware are so-
overhead (the curse of dimensionality eluded to earlier). called hybrid quantum-classical algorithms that partition the
Whereas the storage of classical bits is stable, as the ‘0’ or ‘1’ computational workload between a classical and a quantum
is copied many times over, for instance, in the many spins of a computer.[124,125] The most prominent example of this class of
magnetic patch storing a bit on a hard drive, the situation is algorithms relevant for electronic structure calculations is the
markedly different in quantum computation. Quantum informa- variational quantum eigensolver (VQE), which presents a varia-
tion cannot be copied and is therefore much more fragile and tional (and thus heuristic) algorithm to upper bounding the
affected by quantum noise, the previously mentioned decoher- ground state energy problem.[126–128] In a nutshell, a given wave
ence. Consequently, storage and gate operations on quantum function ansatz is represented with a parametrized quantum
bits are prone to errors that need to be corrected. Note that circuit (i. e. a quantum circuit, where the individual gates are
quantum error correction, first put forward by Shor in 1996,[115] dependent on parameters), and its energy is estimated with a
alone is not sufficient, because the quantum circuits carrying term-by-term measurement.
out the error correction may themselves introduce new errors. As we have discussed, there exist wave function ansätze
Luckily, techniques have been developed that replace each inspired by the ones described in the previous section for which
noisy physical qubit and physical gate in a circuit with an the cost of measurement would grow exponentially on classical
encoded logical qubit and logical gate, thereby allowing an hardware and polynomially on quantum hardware, therefore
entirely noise-free quantum computation to be simulated on allowing to circumvent the exponential overhead. However,
noisy quantum hardware (given that the noise is below a VQE yields the exact solution to the full-CI problem only if the
threshold – the ‘threshold theorem’).[116] In addition to the final wave function can be represented with that ansatz. The
quantum hardware and quantum algorithms, quantum com- parameters of the wave function ansatz are then optimized in
puters therefore also require an important error correction layer an outer loop on a classical computer taking the quantum
in between. That fault-tolerant quantum computation is energy measurements as input in order to obtain a variational
possible at all is mathematically highly non-trivial, but allows us upper bound to the true ground state energy.
to imagine that digital quantum simulation is possible in the Other examples of hybrid classical-quantum algorithms are
future, despite decoherence possibly staying around indefi- the variational quantum time evolution for quantum dynamics
nitely. simulations[129,130] and the variational imaginary-time
There is a price to be paid for the fault-tolerant hardware in evolution,[131] which represents an alternative to VQE for ground
terms of the overhead required. Whereas recently, it has been state energy optimizations.
shown that only a constant overhead is necessary (i. e. the On a fault-tolerant computer, one may directly measure the
number of logical qubits is proportional to the number of full-CI energy by first preparing a wave function ansatz on the
physical ones)[117] and one may therefore generally see no quantum hardware (as in VQE) and then applying the Quantum
fundamental obstacle to building a perfect digital quantum Phase Estimation algorithm, which projects a prepared state
simulator in the future, the number of physical qubits required into an energy eigenstate of the Hamiltonian.[111,132–136] The
1�
for the first viable algorithms to run is extremely large and ground state will be found with high probability after O k
therefore not expected to be reached in the near future, but repetitions, where k is the overlap between the prepared state
only in the medium or long term. We note that one may hope and the exact ground-state wave function. Note, however, that
for novel hardware solutions that lead to a lower overhead due the ground state energy problem is known to be QMA-hard
to stronger protection of the native qubits to noise, e. g., and it might, therefore, be expected that no general exponen-
through topological protection.[118,119] tial speedup will be obtained, rendering this again a heuristic
Therefore, research on quantum algorithms splits, on the algorithm. In practice, however, for those electronic structure
one hand, into algorithms for fault-tolerant quantum com- problems that are dominated by dynamic correlation (see
puters, which we may think of as a future perfect quantum above), the Hartree-Fock state will be a reliable initial state with
computer, and, on the other hand, into algorithms for near- sufficient overlap with the final state, as this has been the key
term or noisy-intermediate stage quantum (NISQ)[120] devices of insight into the development of single-reference approaches
dozens to thousands of physical qubits, that are characterized such as coupled cluster for decades. In the presence of strong
by the absence of complete fault-tolerance. static correlation, the Hartree-Fock state may not have a
Naturally, the algorithm design focuses on rigorous runtime sufficiently large overlap with the final wave function (see Ref.
bounds for the fault-tolerant machines and heuristic arguments [137]). However, in these cases, an amazing qualitative and
for NISQ devices. Common to problems from both chemistry quantitative understanding of good candidates for functions
and molecular biology is the necessity of mapping the (such as the FCI-type wave functions optimized by DMRG or
representation of electrons onto the qubit structure. This FCIQMC; see above) has been assembled that approximate such

ChemBioChem 2023, 24, e202300120 (12 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

strongly correlated systems well so that there is hope that the recognized as a potential high-impact application of future
overlap k indeed can be made substantial. fault-tolerant quantum computers.[140,141] Accordingly, chemical
For Quantum Phase Estimation, one is required to perform a reactions in biomolecular systems are an obvious target for
time evolution with a Hamiltonian that describes the quantum computations with significant advantages under
(bio)molecule under study with parameters that have been certain conditions. In fact, such a scenario has been first
calculated as integrals in a so-called four-index transformation elaborated in detail for the mode of action of the enzyme
from the atomic orbital basis set to the molecular orbital set. nitrogenase[140] and was further expanded on for other catalytic
Since the Hamiltonian consists of a sum of terms, the basic idea processes with improved quantum algorithms.[142]
is to use a Lie-Trotter formula in order to be able to decompose The key feature exploited in this context is the spatial
the generally complicated time evolution into gates that only locality of chemical reactions; i. e., they involve comparatively
touch two particles at a time. For a Hamiltonian with two terms few atoms that are involved in dissociation and association
H ¼ H1 þ H2 time-evolved for time t, this yields the following (bond-breaking and bond-formation) processes. The huge rest
approximation of the biomolecular machinery is treated as a spectator environ-
ment in a given chemical reaction step. This environment can
e itH=�h
¼e itH1 =�h itH2 =�h be efficiently described in a classical embedding scheme, an
� � idea that won the Nobel Prize in Chemistry in 2013.[48] As a
� e iH1 t=N�h e iH2 t=N�h � � � e iH1 t=N�h e iH2 t=N�h ; (5)
result, the complex environment can be efficiently modeled
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
N terms and only the local quantum description of the reaction event
remains the major challenge: the biochemical problem has
whose quality improves for increasing N. Note that it is a been reformulated as a purely chemical problem, and all
sequence of shorter propagations each time only involving one developed solution approaches in the chemical context
of the terms in the Hamiltonian. By now, significant improve- automatically transfer to the biochemical application.
ments on this initial decomposition idea have been devised and In reaction chemistry, these processes are all related to
an excellent, indeed asymptotically optimal, understanding has changing the connectivity of atoms in molecules, i. e., to
been obtained.[138,139] Yet, significant further development is changing covalent bonding in some way. Examples are
needed, especially in regard to concrete and tailored instances enzymatic reactions, but also any kind of chemical reaction that
of simulation problems. carries biological significance. Hence, it starts already with the
Finally, we emphasize that whether analog or digital, fault- creation of the first biomolecules (abiogenesis) and the origin
tolerant or NISQ, it is important to realize that asymptotic of life. Accordingly, quantum computation promises to advance
runtime bounds will only be a guide to the performance and not only the accuracy of quantum predictions for enzyme
practical resource estimates will have to be obtained. These kinetics, especially for strong correlation cases as they are found
bounds will additionally depend strongly on the available in metalloenzymes with 3d metal-atom active sites such as
hardware, e. g. on the number, the connectivity, and the control copper and iron-sulfur enzymes (including electron and proton
of the qubits, on the gate set, and the type of noise. To make transfer processes), but also for stoichiometric reactions and
practical use of quantum computation for molecular biology (or biochemical pathways in living cells.
molecular simulations in general), all of these aspects of the Naturally, any chemical reaction network of relevance to
available hardware need to be combined with an in-depth molecular biology can benefit from quantum computation.
knowledge of the molecular system(s) at hand and, therefore, Intellectually appealing examples may even be taken from
might require highly specialized knowledge and expertise prebiotic chemistry (see, for example, the formose reaction
ranging from molecular biology to the understanding of the network[143,144]), where (traditional and quantum) simulations
state of the quantum hardware to eventually the quantum may provide insights into prebiotic potentialities that have not
algorithm design. been scrutinized by experiment yet. Quantum computation
In the remaining parts of the manuscript, we will detail the might even deliver accurate electronic energy estimates for
prospects of quantum computation in several distinct areas of reactive biomolecular precursors on surfaces that act as
molecular biology. catalysts for the production of the first (functional) biomole-
cules (cf. the Waechtershaeuser hypothesis[145,146]).
Even more relevant would be the application of quantum
computing to simulate reactions that are difficult to reproduce
6. Local quantum effects in biochemical in a laboratory setting. An example are astrochemical reaction
processes networks emerging under the extreme conditions of extra-
terrestrial environments. In the context of molecule formation
We will now turn our attention to (local) quantum mechanical in interstellar space, often the lack of an environment
phenomena in molecular biology, which mostly relate to (producing conditions far away from thermal equilibrium) will
chemical reactions of biomolecules in the most general sense require high-accuracy rovibronic quantum-state resolved for-
(not only covering stoichiometric or enzymatic reactions, but mation theories and corresponding calculations. The high
any kind of chemistry of biomolecules, even involving solvation accuracy needed to reliably predict the outcome of these
and association processes). Chemical processes have been reactions, combined with the relatively small size of the

ChemBioChem 2023, 24, e202300120 (13 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

intermediates entering astrochemical reaction networks, makes at their very heart, governed by dynamic electron correlation
astrochemistry[147–149] an appealing target for quantum comput- effects. While these can often be well modelled by classical
ing. In the context of astrobiology, biomolecules (such as amino surrogate potentials, this is not always the case. Two prominent
acids) on surfaces of interstellar objects (exoplanets or examples are light harvesting and magnetoreception (see Fig-
asteroids) subjected to the light of stars may well fall into the ure 6). Light harvesting represents the initial step of photosyn-
category of approaches applicable to pre-biotic scenarios on thesis, i. e. the process exploited by plants to convert light into
Earth.[150] Moreover, computer simulation can even adjust the chemical energy. Photosynthesis proceeds via a chain of
conditions to account for the macroscopic constraints such as complex chemical processes, the first of which is light harvest-
increased temperature and pressure. Hence, within this context, ing, which corresponds to the absorption of visible light by so-
potential routes leading to the emergence of life in outer space called pigment-protein complexes.[154–156] Magnetoreception[157]
could also be scrutinized based on quantum computing is the process that enables birds to sense weak magnetic fields
simulations. and, in particular, the Earth magnetic field that can be used as a
In the physical description, chemical processes are mapped guide then. It has been argued that quantum effects play a key
to a path across the Born-Oppenheimer PES (cf. Figure 1) so role in both phenomena.
that the calculation of the electronic energy as a solution of the In contrast to the chemical reactions described in the
electronic Schrödinger equation is the key first step (cf. previous section, quantum effects span, in this case, the whole
Figure 2). Accordingly, the accurate quantum mechanical calcu- molecule. These phenomena are the subject of study of so-
lation of the electronic energy for a given arrangement of the called quantum biology.[19,22,158] We prefer not to use this term
atomic nuclei in the biomolecules under consideration suffers because the effects described in Section 6 have also a quantum
from the curse of dimensionality through the large number of nature, but are usually not referred to in the realm of quantum
orbitals required for accurate results. biology.
In the past decades, a zoo of practical solution methods has Pigment-protein complexes encapsulate several chromo-
been devised for traditional computing hardware (see Sec- phores, usually porphyrin derivatives. The light is first absorbed
tion 4). However, every method strikes a compromise between by the chromophores. The resulting electronic excitation is then
feasibility and accuracy: the more accurate the energies (and, in transferred, via a chain of excitation energy transfer (EET)
particular, the energy differences) need to be, the smaller the processes, to the reactive center that converts it into chemical
systems (in terms of the number of atoms or electrons) are for energy.[159,160] The overall efficiency of photosynthesis depends
which an accurate method is feasible on some traditional critically on whether the electromagnetic energy of the
computer hardware. Still, these traditional approaches have absorbed light can be transferred to the reaction center without
reached a remarkable degree of sophistication, accuracy, losses. Multidimensional electronic spectroscopy measurements
usefulness, and acceptance.[151–153] Detailed resource of light-harvesting systems have revealed the presence of
[13,140]
estimates showed that algorithms based on quantum quantum coherences (i. e., that the electronic wave function is a
phase estimation are likely to compete and beat the best coherent superposition of excited states centered on each
traditional approaches for cases of strong electron correlation chromophore) in the early phase of the EET process.[161–163] This
as they occur in active sites of metalloenzymes (and in observation, combined with the fact that quantum mechanical
electronically excited states; see below). Their main advantage models predict a higher efficiency compared to semiclassical
though is the explicit control over the error bar on the result calculations, has led to the hypothesis that light harvesting is
(provided that the initial state prepared features sufficient driven by quantum effects. Although it was later observed that
overlap with the final state; see discussion in the previous the lifetime of electronic coherences may not be long enough
section), which is not available from any traditional approach to have an effect on the EET rate,[164,165] it has been argued that
and also not from the currently most favored algorithm for molecular vibrations, through their coupling with the electronic
NISQ devices, namely, VQE with a unitary coupled cluster ansatz excitation (usually referred to as excitonic coupling), may
because of limitations on the underlying parametrization. enhance the coherence lifetime by quantum coherences.[166–168]
Similar observations have been made for magnetoreception.
According to the currently most favored hypothesis, magneto-
reception relies on the generation of a radical pair in so-called
7. Global quantum effects in functional cryptochromes.[26,27,29] The latter are proteins located, in partic-
biomolecules ular, in the bird’s eyes, where they can absorb visible light. Light
absorption triggers a chain of electron transfer reactions that
We highlighted in the previous section that a key feature of generates two unpaired electrons (the radical pair) localized on
quantum effects associated with the rearrangement of nuclei a flavin adenine dinucleotide and a tryptophan residue of the
and electrons that drive biochemical reactions is their locality in protein. The spin state of the radical pair can be either singlet
space – i. e. they are relevant only for the reacting nuclei and or triplet, and an external magnetic field can affect the relative
their electrons (and for a comparatively tiny fraction of their amount of these two states. Since the rate at which the radical
environment). Other biomolecular processes are, instead, pair decays to the electronic ground state via spin recombina-
influenced by quantum effects acting on a larger spatial scale. tion depends on the spin state, a magnetic field ultimately
For instance, large amplitude motions of solvated proteins are, influences the reaction outcome. This magnetic field sensitivity

ChemBioChem 2023, 24, e202300120 (14 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

Figure 6. Key steps involved in the modelling of photosynthesis (left part) and magnetoreception (right part; “Biological function”). First, the biomolecule that
tunes the target biological function is identified (“Biochemical system”): a light- harvesting complex for photosynthesis and a cryptochrome protein for
magnetoreception. Second, the “Physical process” linked to the biological function of the molecules is identified. For photosynthesis, this is the excitation
energy transfer (EET) that enables to funnel the energy captured by the chromophores through light absorption to the reaction center. For magnetoreception,
the key physical process is the generation of the radical pair in the FAD-tryptophan complex inside the cryptochrome. Finally, one defines a “Model
Hamiltonian” that describes the quantum process which tunes the biochemical function. For the EET in a light-harvesting complex, the shown excitonic
Hamiltonian can describe the chromophores as two-level quantum systems (with energies ɛ0,i and ɛ1,i for the i-th chromophore, both of which depend on the
chromophore nuclear coordinates Qi), that interact via a nearest-neighbor potential. For the radical pair dynamics, each electron of the radical pair is
described by two spin operators, Ŝ1 and Ŝ2 that interact with the Earth magnetic field B, among themselves via a dipolar spin-spin interaction (D is the dipolar
interaction tensor, whereas I is the unit matrix), and with the nuclear spin (referred to as Îk for the k-th nucleus). Note that a system as large as the
cryptochrome contains several dozens of nuclei with non-zero spin.

is believed to be at the origin of mag netoreception.[169,170] thousands of electrons, and this makes applying the theory
Semiclassical simulations predict a radical pair lifetime that is described in Section 3 extremely challenging for both classical
much shorter than that obtained with quantum-mechanical and quantum computations. As a way out, effective Hamilto-
models.[171] As for light harvesting, this has been taken as an nians (usually referred to as excitonic Hamiltonians) have been
indication that quantum mechanical effects play a key role in designed that largely reduce the problem complexity but still
magnetoreception. encode the key quantum mechanical effects that drive these
For both phenomena, quantum mechanical effects affect biological phenomena.
the interaction between molecules – the chromophores, for In these Hamiltonians, the chromophores are approximated
light-harvesting, and the unpaired electrons, for magneto- as quasiparticles, the excitons, with two quantum levels each,
reception. Simulating quantum effects happening on such large one corresponding to the electronic ground state and the other
spatial scale is much harder than for chemical reactions. The one to the excited state. The inter-chromophore interaction is
electronic wave functions of these systems depend on then approximated with effective potential energy terms that

ChemBioChem 2023, 24, e202300120 (15 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

account for the excitation energy transfer phenomenon. biomolecules. Although these interactions can be described
Condensing all the electrons of a chromophore into a two-level with the quantum approaches presented in Section 5, such a
quantum system drastically reduces the Hamiltonian dimen- route is both impractical and superfluous. It is impractical
sionality and, consequently, the complexity of solving the because, due to the curse of dimensionality, full atomistic
corresponding quantum equations, but at a huge loss in terms quantum mechanical simulations of large biomolecules will
of detail and resolution. Excitonic Hamiltonians can be further most likely remain out of reach for any simulation method,
extended to include also the nuclear degrees of freedom, which both with classical and quantum computation. Furthermore, it
leads to so-called (phenomenological) vibronic Hamiltonians. is superfluous, because weak interactions are effectively
Even within the excitonic approximation, solving exactly the approximated by classical surrogate potentials, i. e., by the so-
Schrödinger equation for a faithful Hamiltonians describing a called force fields.[177–180] A force field approximates the PES of a
pigment-protein complex or a radical pair is extremely hard molecule through a classical parametric interaction potential
with classical computing algorithms. In the case of light between its atoms, as shown in Figure 1. The interaction energy
harvesting, a pigment-protein complex may be composed of between atoms that are linked by a covalent bond is expressed
more than 10 chromophores, each one with several dozens of as a power series in terms of bond lengths ri and angles θj, and
degrees of freedom. For magnetoreception, the radical pair as a Fourier series in terms of the dihedral angles γm (see
interacts with the nuclear spins. The number of nuclei with non- Figure 1). The non-covalent dispersion-repulsion interaction
zero spin may be large, which makes simulating the radical pair between non-bonded atoms is, instead, expressed as a sum of
dynamics computationally hard. In both cases, this results in pairwise Lennard-Jones-type interactions. The parameters enter-
many-body Hamiltonians depending on more than 100 par- ing the force field (i. e., kiðbÞ , kjðaÞ , kmðdÞ , C6ij , and C12
ij
in Figure 1) are
ticles, which is out of the reach of the classical simulation obtained by least-squares fitting either to results of quantum
strategies described in Section 3. For this reason, the claim that chemical calculations or to experimental data. The force field
global quantum effects are relevant in photosynthesis and then defines a potential that can be used in classical atomistic
magnetoreception relies on simulations of simplified problems simulations based on the Newton equation of motion of the
that further approximate the biochemical process. For instance, atoms.
quantum mechanical simulations of magnetoreception have At this point, it is important to highlight that any biological
been limited to radical pairs that are much smaller and simpler process cannot be considered an isolated process. In fact, it
than the cryptochromes.[172–174] Quantum computing might have happens at about room temperature and involves a continuous
the potential to overcome these limits, provided that either a energy exchange with the environment (for instance, the cell in
sufficiently large number of logical qubits can be provided in which a protein is located). A biomolecule is then not to be
huge fault-tolerant quantum computers or a large analog represented by a single molecular structure, but rather as a
quantum simulator can be built that can represent Hamilto- thermal ensemble of many configurations, each one weighted
nians with so many quasiparticles. by the appropriate thermodynamic weight. The key quantity is,
In fact, both the efficiency of the EET process in pigment- therefore, not the energy of the microstate of a single
protein complexes and the lifetime of the radical pair can be configuration, but the free energy calculated from the partition
predicted by solving the time-dependent Schrödinger equation function of all microstates. The partition function also gives
for the corresponding Hamiltonian. As we discussed in Section access to all thermodynamic properties. The PES is not the
(5), this is one of the few examples of a problem for which appropriate quantity to interpret the reactivity within such a
theoretical guarantees of an exponential speedup can be thermodynamic ensemble as the driving force of a reaction is,
derived. In practice, the major challenge will be ensuring in fact, the free energy (see Figure 7). Typically, the potential of
quantum hardware stability over the whole propagation time. mean forces[181] is then considered, which generalizes the
As we display Figure 5, achieving such stability is particularly concept of PES to thermal systems. For a given reaction profile
hard in digital quantum computing. Conversely, quantum (e. g., one of the paths of Figure 1), the potential of mean forces
simulators may be better suited for this class of simulations. As is calculated, at each point of the PES, by averaging out all
a matter of fact, some pilot experimental realizations of degrees of freedom of the molecule but the reactive one.
quantum simulators for light-harvesting systems have been Despite this conceptual difference, the electronic energies
proposed in recent years,[175,176] although they are limited to remain the cornerstone also for calculating the potential of
very simple excitonic Hamiltonians. mean forces because they are a key contribution to the
partition function and, therefore, to any thermodynamic
quantity. One may wonder whether quantum computing may
8. Quantum computing for structural biology be applied to speed-up also this step. While algorithms to
calculate the partition function of quantum systems on
Processes driven by quantum effects induced either by the quantum hardware have been proposed,[182] which is not
breaking or forming of covalent bonds (see Section 6) or by surprising in view of the quantum simulates quantum principle
light excitation (see Section 7) represent only a very small introduced above, extensions of these schemes to compute
portion of the realm of biological processes. In fact, we already classical thermal averages are still scarce. Quantum computing
highlighted in the introduction that molecular biology is realizations of the Monte-Carlo algorithm exist,[183] and so it may
governed by reversible, non-covalent interactions between be argued that they can be applied to calculate thermal

ChemBioChem 2023, 24, e202300120 (16 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

Figure 7. Graphical representation of the steps needed to calculate the potential of mean force for a biochemical process. The process coordinate s describing,
for instance, the binding of a small molecule (represented as a red circle) to a protein (represented as a green object) may be some distance between these
two molecular systems. The contribution of all the 3Ns 7 remaining coordinates (where Ns is the number of atoms in the molecular system comprised of the
red and the green molecule) is averaged out by Boltzmann weighting in a canonical (N, V, T) ensemble. For the single molecular system, this is represented by
the integral reported in the lower part of the figure, yielding the classical partition function for the molecular system. The potential energy entering the
Boltzmann factor is the electronic energy (i. e., the Born-Oppenheimer hypersurface) or its approximation by a classical surrogate potential defined in Figure 1.
The logarithm of the integral of the classical partition function is proportional to the (molecular) potential of mean force As. The Helmholtz free energy A,
which depends on the system volume V, its temperature T, and the macroscopic number of molecular systems N, is obtained by integrating over each s in
every molecular system in the macroscopic system. The free-energy difference between reactants and products ΔA determines whether a given process is
thermodynamically favorable, while the transition-state free-energy difference ΔA� determines the corresponding rate constant.

averages via sampling techniques.[184–186] However, these algo- possible, for instance, to parametrize or refine a force field that
rithms assume that the distribution function can be encoded as describes molecular interactions in proteins based on exact
the probability distribution of a many-body wave function (see quantum chemical simulations on small polypeptides. A small
Section 3). Hence, applying them to calculate classical thermo- polypeptide may contain a few dozen atoms and a few hundred
dynamic averages requires mapping a classical probability electrons and is, therefore, too large to be targeted by classical
distribution onto the qubit states, a task that is hard itself, as quantum chemical calculations. Such a molecular size may be,
we will discuss in the following section. in the near future, within the application range of quantum
Once an appropriate force field is available, simulating the computing, in which case such a route would provide very
classical dynamics of a biomolecule represents the cornerstone accurate data for parametrizing force fields (provided that the
of several types of biological simulations. The most prominent quantum machine can represent the discretized problem well).
example is the protein folding problem, i. e. the task of This idea has been recently explored in Refs. [190] and [191],
predicting the three-dimensional structure of a protein based where VQE-based methods delivered data for parametrizing
on its amino acid sequence. The protein folding problem can classical force fields. In the future, the very same problems can
be solved by simulating the dynamics of a protein under be explored with algorithms for fault-tolerant quantum com-
thermal equilibrium conditions. For sufficiently long propaga- puters.
tion times, the protein will evolve into its most stable Quantum computing can also be applied to classical
conformation, thereby yielding the solution to the protein simulations that rely on a given force field. For instance, protein
folding problem. The three-dimensional structure of a protein folding can be rephrased as an optimization problem. The
can be reliably predicted with force fields because it is mostly stable structure of a protein corresponds, in fact, to the
determined by the non-covalent interactions between the structure yielding the lowest free energy. However, this
amino acid residues of the protein. Not being affected by the optimization problem is classically complex (NP-hard) to
curse of dimensionality, classical simulations can target much solve.[192] For this reason, the folding problem can be solved
larger molecules compared to quantum mechanical simulations based on this idea only for simplified lattice models that
so that folding simulations of small fast-folding proteins are drastically reduce the conformational freedom of a molecule.[193]
possible.[187–189] Note that protein folding is not the only molecular biology
Since classical simulations are not impacted by the curse of problem that can be rephrased as an optimization problem.
dimensionality, one may argue whether quantum computing Another example is molecular docking, i. e. the prediction of the
can be beneficial also to this field. The first, natural combination most stable binding configuration of a ligand to a given target,
of these two fields is to parametrize a force field based on which represents the first step of in silico drug design. However,
energies obtained from quantum computations. It would be also in this case, the optimal configuration may be obtained by

ChemBioChem 2023, 24, e202300120 (17 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

minimizing a classical force field describing the interaction procedure does not provide any insights into the physical
between the ligand and the target. As for the protein folding, dynamics of the folding process itself. As illustrated by the
the resulting optimization problem is classically hard.[194,195] Levinthal paradox,[208] a protein cannot fold simply by randomly
Since optimization problems may be one of the first and sampling all its possible conformations because, in this way, the
primary application fields of quantum computing, the question folding time would be much higher than what is observed
arises of whether quantum computing may boost the efficiency experimentally. The local interactions between amino acids
of optimization-based molecular biology algorithms. Whereas drive, while the protein folds, the protein toward the lowest-
polynomial-time quantum algorithms are not expected for NP- energy conformation, as highlighted by the famous “folding
hard problems,[196] like general optimization problems, more funnel” metaphor. While it is obvious that a quantum mechan-
modest speedups, still significant in practice, might be achiev- ical description of these interactions to yield accurate electronic
able. An example is the quadratic speedup that can be achieved energies that then enter sampling procedures of structural
based on an exhaustive search for the optimal solution with configurations would be desirable, it is highly uncertain
Grover’s search algorithm[197] (leaving aside questions of whether a quantum machine can accomplish this for such large
efficient data transfer and storage in the quantum machine; see atomistic structures (let alone the fact that also the role of
below). chaperone proteins and other mechanisms that can assist or
Moreover, quantum optimization algorithms for which a steer a folding process may need to be considered). However,
gain in performance cannot be guaranteed, but has been embedding approaches that dissect a quantum system into
observed heuristically for small test cases, have been designed. smaller quantum subsystems, which are then amenable to
In the latter category, the quantum approximate optimization actual computations, can deliver key concepts to meet this
algorithm (QAOA)[198] is noteworthy and is currently under in- challenge.[209–212]
depth investigation.[199–202] Another example of heuristic algo- Finally, we note that a third strategy to solve the protein
rithms is quantum annealing, which solves a quadratic binary folding problem is offered by data-driven approaches. The
optimization by mapping it onto the equivalent problem of three-dimensional structure of many proteins is often available
calculating the ground state configuration of an Ising Hamil- from nuclear magnetic resonance or X-ray crystallography. It is,
tonian. The latter is then calculated on a specialized simulator. therefore, possible to use this enormous amount of experimen-
The idea is that quantum tunneling will explore the compli- tal data as a training set for a machine-learning model. A prime
cated energy landscape of this model more efficiently and, example of such an application is the deep neural network
therefore, avoid converging into local minima.[203,204] AlphaFold.[213] AlphaFold is a specific machine learning algo-
All these classes of optimization algorithms can be, in rithm that can predict the three-dimensional structure of a
principle, applied to biochemical problems such as those protein using, as input, its amino acid sequence. Data-driven
mentioned above. The first attempts to solve the protein approaches can overcome the steep scaling of explicit classical
folding optimization problem relied on quantum and quantum simulations. However, since the former make
annealing[205,206] while, more recently, hybrid classical-variational predictions based on previously learned data, the accuracy of
algorithms based on QAOA,[109] VQE,[110] and on quantum machine learning for genuinely new instances depends on its
random walks[207] have been proposed. The efficiency of all transferability to these unseen input data – a property that is
these methods has only been explored so far for simplified usually referred to as generalization power, which is extremely
lattice models of small polypeptides with a few dozen of amino hard to estimate a priori.
acids. On the one hand, this may be taken as an indication that It is therefore not surprising that it was recently shown[214]
these quantum algorithms may be promising candidates for that data-driven approaches can hardly predict the structure of
molecular biology applications. At the same time, it is so-called intrinsically disordered proteins,[215] in which only a
impossible to assess, based only on this observation, if and small fraction of the amino acid sequence is organized into a
which one of them may yield a practical quantum advantage well-defined three-dimensional structure, while the rest of the
for systems that are out of the reach of classical computers. protein is inherently dynamical and disordered. Inferring the
Furthermore, going beyond simplified lattice models and structure of these proteins based on data-driven approaches is
applying them to a real force field is a crucial step in order to extremely challenging because their conformation is inherently
truly compete with methods based on classical dynamics. We different from any other protein and inference from data
note that finding the lowest-energy conformation of a polypep- lacking this information breaks down. Physical modeling such
tide is not the unique relevant optimization problem in protein as first-principles simulations does not suffer from these
science. For instance, it was proposed in Ref. [204] to limitations by construction, but many practical challenges of
reformulate the problem finding the most favourable reaction simulating intrinsically disordered proteins with classical algo-
pathways as an optimization problem, and to solve it with rithms remain.[216] We will discuss in the next section whether
quantum-annealing algorithms. quantum computing may have the potential to also advance
A key conceptual difference between simulation- and machine learning applications.
optimization-based solutions to the protein folding problem
will remain for both classical and quantum computing. Solving
folding as an optimization problem yields only the final
structure, corresponding to the lowest (free) energy, but this

ChemBioChem 2023, 24, e202300120 (18 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

9. Quantum computing for data-driven if the terminal segments of their nucleobase sequence overlap,
approaches to molecular biology genome sequencing becomes equivalent to the problem of
finding a Hamiltonian cycle of the graph. This problem is known
to be NP-hard classically,[233,234] although efficient heuristic
One may argue whether quantum computing has the potential classical solutions to it have been developed over the
to boost machine-learning algorithms in molecular years.[235,236] The heuristic quantum algorithms introduced in
biology.[217,218] Answering this question is the overarching goal Section 8 for solving optimization problems can be straightfor-
of quantum machine learning.[219,220] In essence, machine learn- wardly applied also to this context.
ing is a mathematical framework to learn a target function by A key take-home message of the discussion so far is that
knowing its value at a given set of points (usually known as quantum machine learning may become applicable, in the
training set). The target function is approximated with a future, only for data-driven quantum algorithms in which a
parameterized function, and the parameters are optimized to quantum speedup can be observed already for small data sets.
reproduce its value on the training set. Otherwise, either loading the classical data on quantum
Quantum computing may boost machine learning in two hardware or carrying out the classical optimization will become
respects. First, it can speed up the training process. Second, it the main bottleneck, nullifying any quantum advantage. For
can make the algorithm more expressive in order to be able to instance, quantum machine learning algorithms for identifying
predict more complex functions. The first advantage has been common patterns between two DNA sequences – a task known
leveraged in deriving the quantum realization of the principal as DNA alignment – have been proposed in recent years.[237,238]
components analysis algorithm[221] and of a support vector As already mentioned above, the Grover search yields only a
machine.[222] Formal guarantees of their quantum speedup over quadratic speedup over its classical realization. With such a
their classical counterpart might be available for these limited speedup, a practical advantage will only be observed for
algorithms, but they require, as the initial step, preparing the problems with very large data sets.[239] However, these will be
qubits in a linear superposition with coefficients that depend difficult to deal with on currently designed hardware with
on the input data set. In principle, such a step can be realized rather few physical qubits (to be compared to the vast number
with the quantum realization of the random access memory of bits in a classical computer). Moreover, it remains to be seen
(RAM) architecture.[223] However, depending on the nature of how to experimentally realize a so-called quantum random
the input data, this loading phase may become the bottleneck access memory,[223] which would represent a crucial component
of the overall simulation, therefore eliminating any practical for loading a large data set on a quantum computer. For these
quantum advantage.[224] reasons, one may not expect quantum algorithms based on the
Methods that leverage the quantum computer only for Grover search to represent – at least in the foreseeable future –
boosting the representation power of the machine-learning suitable candidates for demonstrating a practical quantum
model do not suffer, in principle, from this limitation because advantage.
they can carry out the training process classically. For instance, Other applications of quantum machine learning algorithms
a neural network approximates the target functions in terms of may, in principle, yield a practical quantum advantage in the
elementary, non-linear functions – the so-called neurons. future. This is the case of the recently proposed quantum
Increasing the number of neurons increases the representation algorithm for DNA sequencing[240,241] and for identifying the
power of the network but, at the same time, makes the training most probable binding site of a transcription factor to DNA.[242]
phase more complex. A quantum circuit (see Figure 5) can be In both cases, the quantum hardware is leveraged to solve a
interpreted as a quantum neural network, where the gates are classically hard optimization problem with heuristic algorithms.
the quantum counterpart of the neurons. For a given number However, being heuristic, the efficiency of these algorithms
of neurons, quantum neural networks can be more expressive – must be assessed at a case-by-case level. At this stage of
even exponentially, for specific network architectures – than hardware development, with quantum computers composed of
their classical counterpart.[225–227] Note that this expressivity at most hundreds of noisy qubits (see Section 5), these
boost is not limited to neural networks and has been observed, algorithms can be validated only on very small data sets. It is,
for instance, for support vector machines.[222,228,229] However, the therefore, currently extremely hard to predict whether they will
training phase, when carried out classically, may itself become yield an advantage when applied to problems that are out of
hard due to the emergence of so-called Barren plateaus, the reach of classical hardware. This limitation does not only
especially for large data sets.[230,231] This makes the training hold for machine learning algorithms, but for any problem
process itself a computationally hard problem and therefore where classical data are mapped onto the qubit space,
eradicates a quantum speedup. including, among others, the task of finding the lowest-energy
Quantum algorithms for solving data-driven problems are configuration of a biological macromolecule given a force field
not limited to machine learning. These algorithms can also be describing the interaction between its atoms.
used to identify correlations within data that are collected
experimentally. For instance, genome sequencing requires
reconstructing the nucleobase sequence of an organism based
on small portions of the full DNA sequence.[232] By interpreting
the DNA segments as vertices of a graph, which are connected

ChemBioChem 2023, 24, e202300120 (19 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

10. Conclusions other branches of molecular and materials science. Accordingly,

approaches for modelling biochemical processes will also allow
The living world is built from atoms and molecules and, for modelling of general chemical reactions. However, this is
therefore, it is clearly governed by quantum physics. Due to the true also from a more general perspective of physical modelling
large size of most biomolecules and the relatively high temper- or data-driven science. For instance, approaches for protein
atures of living organisms, classicality emerges via the process simulation will also be applicable in the context of polymer
of decoherence. This raises the question in which situations physics.
accurate quantum mechanical modeling is required and where
effective classical description suffices. In an in-silico descriptive
and predictive context, this question directly affects the Acknowledgments
computational resources required to simulate biological proc-
esses, because full quantum mechanical simulations are The authors are grateful for the generous funding through the
affected by the curse of dimensionality when simulated with ‘Quantum for Life Center’ funded by the Novo Nordisk Foundation
classical computers. (grant NNF20OC0059939). Open Access funding provided by
This article has laid out the prospects of quantum comput- Eidgenössische Technische Hochschule Zürich.
ing coming to rescue, moving from smaller to larger biomo-
lecular structures. Quantum mechanical simulations are essen-
tial when simulating biochemical processes that are Conflict of Interests
accompanied by the creation of new chemical bonds. Simulat-
ing these phenomena has, therefore, the highest prospect of The authors declare no conflict of interest.
impact by quantum computers. However, we also discussed
how quantum effects may be crucial also in larger functional
biomolecules, where quantum effects shape classical effective Data Availability Statement
descriptions and require a more targeted use of quantum
computing at this quantum-classical barrier. Data sharing is not applicable to this article as no new data
Moving to even larger structures with an effectively cured were created or analyzed in this study.
curse of dimensionality, biocomputations are now faced with
solving a still hard, but now entirely classical, computational Keywords: Quantum Computing · Molecular Biology ·
problem, often in the form of an optimization problem, e. g. in Quantum Chemistry · Quantum Biology · Molecular Simulations
protein folding. Also here, quantum computing might come to
supplement classical computers. From a proven quadratic
Grover speedup for exhaustive search strategies to tailored [1] J. Drenth, Principles of Protein X-Ray Christallography, Springer New
algorithms to be run on fault-tolerant computers, to the York, NY 2007.
heuristic approaches for near-term quantum devices, there is a [2] M. M. Seibert, T. Ekeberg, F. R. N. C. Maia, M. Svenda, J. Andreasson, O.
Jönsson, D. Odić, B. Iwan, A. Rocker, D. Westphal, M. Hantke, D. P.
long list of approaches for this class of problems, which appears DePonte, A. Barty, J. Schulz, L. Gumprecht, N. Coppola, A. Aquila, M.
not only in biomolecular sciences but in all sciences and Liang, T. A. White, A. Martin, C. Caleman, S. Stern, C. Abergel, i. Seltzer,
industries, ranging from logistics to finance. J.-M. Claverie, C. Bostedt, J. D. Bozek, S. Boutet, A. A. Miahnahri, M.
Messerschmidt, J. Krzywinski, G. Williams, K. O. Hodgson, M. J. Bogan,
Nevertheless, the specifics of the optimization problems in C. Y. Hampton, R. G. Sierra, D. Starodub, I. Andersson, S. Bajt, M.
biomolecular sciences require a targeted and in-depth under- Barthelmess, J. C. H. Spence, P. Fromme, U. Weierstall, R. Kirian, M.
standing of the type of optimization problem as well as a Hunter, R. B. Doak, S. Marchesini, S. P. Hau-Riege, M. Frank, R. L.
Shoeman, L. Lomb, S. W. Epp, R. Hartmann, D. Rolles, A. Rudenko, C.
tailored solution. The same is true more generally for important Schmidt, L. Foucar, N. Kimmel, P. Holl, B. Rudek, B. Erk, A. Hömke, C.
classical computational problems in the life sciences, ranging Reich, D. Pietschner, G. Weidenspointner, L. Strüder, G. Hauser, H.
from genomics to patient data from, e. g., imaging or sensory Gorke, J. Ullrich, I. Schlichting, S. Herrmann, G. Schaller, F. Schopper, H.
Soltau, K.-U. Kühnel, R. Andritschke, C.-D. Schröter, F. Krasniqi, M. Bott,
data such as from wearable devices. Here, it seems possible that S. Schorb, D. Rupp, M. Adolph, T. Gorkhover, H. Hirsemann, G.
quantum computers might become a supplement and accel- Potdevin, H. Graafsma, B. Nilsson, H. N. Chapman, J. Hajdu, Nature
erator of machine learning methods. More generally, quantum 2011, 470, 78.
[3] H. N. Chapman, P. Fromme, A. Barty, T. A. White, R. A. Kirian, A. Aquila,
technologies, such as quantum sensing,[243] might enhance the M. S. Hunter, J. Schulz, D. P. DePonte, U. Weierstall, R. B. Doak,
quality of imaging data of biomolecular structures (see e. g. F. R. N. C. Maia, A. V. Martin, I. Schlichting, L. Lomb, N. Coppola, R. L.
Ref. [244]) and, therefore, improve the quality of the data on Shoeman, S. W. Epp, R. Hartmann, D. Rolles, A. Rudenko, L. Foucar, N.
Kimmel, G. Weidenspointner, P. Holl, M. Liang, M. Barthelmess, C.
which further classical or quantum computations would be Caleman, S. Boutet, M. J. Bogan, J. Krzywinski, C. Bostedt, S. Bajt, L.
based. In a future scenario, not only the quantum computers Gumprecht, B. Rudek, B. Erk, C. Schmidt, A. Hömke, C. Reich, D.
but also the quantum sensors might be connected with help of Pietschner, L. Strüder, G. Hauser, H. Gorke, J. Ullrich, S. Herrmann, G.
Schaller, F. Schopper, H. Soltau, K.-U. Kühnel, M. Messerschmidt, J. D.
entanglement, thereby harnessing a further increase in reso- Bozek, S. P. Hau-Riege, M. Frank, C. Y. Hampton, R. G. Sierra, D.
lution (cf. Ref. [245]). Starodub, G. J. Williams, J. Hajdu, N. Timneanu, M. M. Seibert, J.
We emphasize that the first-principles nature of the Andreasson, A. Rocker, O. Jönsson, M. Svenda, S. Stern, K. Nass, R.
Andritschke, C.-D. Schröter, F. Krasniqi, M. Bott, K. E. Schmidt, X. Wang,
quantum mechanical basis of the physical modelling approach I. Grotjohann, J. M. Holton, T. R. M. Barends, R. Neutze, S. Marchesini, R.
makes the methods discussed in this work also applicable to Fromme, S. Schorb, D. Rupp, M. Adolph, T. Gorkhover, I. Andersson, H.

ChemBioChem 2023, 24, e202300120 (20 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

Hirsemann, G. Potdevin, H. Graafsma, B. Nilsson, J. C. H. Spence, Nature Curry, C. Hamilton, E. Sodergren, D. A. Wheeler, W. Barris, G. L. Bennett,
2011, 470, 73. A. Eggen, R. D. Green, G. P. Harhay, M. Hobbs, O. Jann, J. W. Keele,
[4] J. C. H. Spence, U. Weierstall, H. N. Chapman, Rep. Prog. Phys. 2012, 75, M. P. Kent, S. Lien, S. D. McKay, S. McWilliam, A. Ratnakumar, R. D.
102601. Schnabel, T. Smith, W. M. Snelling, T. S. Sonstegard, R. T. Stone, Y.
[5] A. T. Chinwalla, L. L. Cook, K. D. Delehaunty, G. A. Fewell, L. A. Fulton, Sugimoto, A. Takasuga, J. F. Taylor, C. P. V. Tassell, M. D. MacNeil, A. R.
R. S. Fulton, T. A. Graves, L. W. Hillier, E. R. Mardis, J. D. McPherson, T. L. Abatepaulo, C. A. Abbey, V. Ahola, I. G. Almeida, A. F. Amadio, E.
Miner, W. E. Nash, J. O. Nelson, M. N. Nhan, K. H. Pepin, C. S. Pohl, T. C. Anatriello, S. M. Bahadue, F. H. Biase, C. R. Boldt, J. A. Carroll, W. A.
Ponce, B. Schultz, J. Thompson, E. Trevaskis, R. H. Waterston, M. C. Carvalho, E. P. Cervelatti, E. Chacko, J. E. Chapin, Y. Cheng, J. Choi, A. J.
Wendl, R. K. Wilson, S.-P. Yang, P. An, E. Berry, B. Birren, T. Bloom, D. G. Colley, T. A. de Campos, M. D. Donato, I. K. de Miranda Santos, C. J.
Brown, J. Butler, M. Daly, R. David, J. Deri, S. Dodge, K. Foley, D. Gage, de Oliveira, H. Deobald, E. Devinoy, K. E. Donohue, P. Dovc, A. Eberlein,
S. Gnerre, T. Holzer, D. B. Jaffe, M. Kamal, E. K. Karlsson, C. Kells, A. C. J. Fitzsimmons, A. M. Franzin, G. R. Garcia, S. Genini, C. J. Gladney,
Kirby, E. J. Kulbokas, E. S. Lander, T. Landers, J. P. Leger, R. Levine, K. J. R. Grant, M. L. Greaser, J. A. Green, D. L. Hadsell, H. A. Hakimov, R.
Lindblad-Toh, E. Mauceli, J. H. Mayer, M. McCarthy, J. Meldrim, J. P. Halgren, J. L. Harrow, E. A. Hart, N. Hastings, M. Hernandez, Z.-L. Hu, A.
Mesirov, R. Nicol, C. Nusbaum, S. Seaman, T. Sharpe, A. Sheridan, J. B. Ingham, T. Iso-Touru, C. Jamis, K. Jensen, D. Kapetis, T. Kerr, S. S. Khalil,
Singer, R. Santos, B. Spencer, N. Stange-Thomann, J. P. Vinson, C. M. H. Khatib, D. Kolbehdari, C. G. Kumar, D. Kumar, R. Leach, J. C.-M. Lee,
Wade, J. Wierzbowski, D. Wyman, M. C. Zody, E. Birney, N. Goldman, A. C. Li, K. M. Logan, R. Malinverni, E. Marques, W. F. Martin, N. F. Martins,
Kasprzyk, E. Mongin, A. G. Rust, G. Slater, A. Stabenau, A. Ureta-Vidal, S. S. R. Maruyama, R. Mazza, K. L. McLean, J. F. Medrano, B. T. Moreno,
Whelan, R. Ainscough, J. Attwood, J. Bailey, K. Barlow, S. Beck, J. D. D. Moré, C. T. Muntean, H. P. Nandakumar, M. F. Nogueira, I. Olsaker,
Burton, M. Clamp, C. Clee, A. Coulson, J. Cuff, V. Curwen, T. Cutts, J. S. D. Pant, F. Panzitta, R. C. Pastor, M. A. Poli, N. Poslusny, S. Rachagani,
Davies, E. Eyras, D. Grafham, S. Gregory, T. Hubbard, A. Hunt, M. Jones, S. Ranganathan, A. Razpet, P. K. Riggs, G. Rincon, N. Rodriguez-Osorio,
A. Joy, S. Leonard, C. Lloyd, L. Matthews, S. McLaren, K. McLay, B. S. L. Rodriguez-Zas, N. E. Romero, A. Rosenwald, L. Sando, S. M.
Meredith, J. C. Mullikin, Z. Ning, K. Oliver, E. Overton-Larty, R. Plumb, S. Schmutz, L. Shen, L. Sherman, B. R. Southey, Y. S. Lutzow, J. V.
Potter, M. Quail, J. Rogers, C. Scott, S. Searle, R. Shownkeen, S. Sims, M. Sweedler, I. Tammen, B. P. V. Telugu, J. M. Urbanski, Y. T. Utsunomiya,
Wall, A. P. West, D. Willey, S. Williams, J. F. Abril, R. Guigó, G. Parra, P. C. P. Verschoor, A. J. Waardenberg, Z. Wang, R. Ward, R. Weikard, T. H.
Agarwal, R. Agarwala, D. M. Church, W. Hlavina, D. R. Maglott, V. Welsh, S. N. White, L. G. Wilming, K. R. Wunderlich, J. Yang, F.-Q. Zhao,
Sapojnikov, M. Alexandersson, L. Pachter, S. E. Antonarakis, E. T. Science 2009, 324, 522.
Dermitzakis, A. Reymond, C. Ucla, R. Baertsch, M. Diekhans, T. S. Furey, [8] E. R. Mardis, Nature 2011, 470, 198.
A. Hinrichs, F. Hsu, D. Karolchik, W. J. Kent, K. M. Roskin, M. S. Schwartz, [9] I. Kassal, J. D. Whitfield, A. Perdomo-Ortiz, M.-H. Yung, A. Aspuru-Guzik,
C. Sugnet, R. J. Weber, P. Bork, I. Letunic, M. Suyama, D. Torrents, E. M. Ann. Rev. Phys. Chem. 2011, 62, 185.
Zdobnov, M. Botcherby, S. D. Brown, R. D. Campbell, I. Jackson, N. Bray, [10] Y. Cao, J. Romero, J. P. Olson, M. Degroote, P. D. Johnson, M. Kieferová,
O. Couronne, I. Dubchak, A. Poliakov, E. M. Rubin, M. R. Brent, P. Flicek, I. D. Kivlichan, T. Menke, B. Peropadre, N. P. D. Sawaya, S. Sim, L. Veis,
E. Keibler, I. Korf, S. Batalov, C. Bult, W. N. Frankel, P. Carninci, Y.
A. Aspuru-Guzik, Chem. Rev. 2019, 119, 10856.
Hayashizaki, J. Kawai, Y. Okazaki, S. Cawley, D. Kulp, R. Wheeler, F.
[11] B. Bauer, S. Bravyi, M. Motta, G. K.-L. Chan, Chem. Rev. 2020, 120,
Chiaromonte, F. S. Collins, A. Felsenfeld, M. Guyer, J. Peterson, K.
12685.
Wetterstrand, R. R. Copley, R. Mott, C. Dewey, N. J. Dickens, R. D. Emes,
[12] S. McArdle, S. Endo, A. Aspuru-Guzik, S. C. Benjamin, X. Yuan, Rev. Mod.
L. Goodstadt, C. P. Ponting, E. Winter, D. M. Dunn, A. C. von Niederhau-
Phys. 2020, 92, 015003.
sern, R. B. Weiss, S. R. Eddy, L. S. Johnson, T. A. Jones, L. Elnitski, D. L.
[13] H. Liu, G. H. Low, D. S. Steiger, T. Häner, M. Reiher, M. Troyer, Matherials
Kolbe, P. Eswara, W. Miller, M. J. O’Connor, S. Schwartz, R. A. Gibbs,
Theory 2022, 6, 11.
D. M. Muzny, G. Glusman, A. Smit, E. D. Green, R. C. Hardison, S. Yang,
[14] M. Reiher (Editor), Atomistic approaches in modern biology, Topics in
D. Haussler, A. Hua, B. A. Roe, R. S. Kucherlapati, K. T. Montgomery, J.
Current Chemistry, Springer, Berlin, Germany, Volume 268 2007.
Li, M. Li, S. Lucas, B. Ma, W. R. McCombie, M. Morgan, P. Pevzner, G.
[15] C. Outeiral, M. Strahm, J. Shi, G. M. Morris, S. C. Benjamin, C. M. Deane,
Tesler, J. Schultz, D. R. Smith, J. Tromp, K. C. Worley, Nature 2002, 420,
WIREs Comput. Mol. Sci. 2021, 11, e1481.
520.
[16] P. S. Emani, J. Warrell, A. Anticevic, S. Bekiranov, M. Gandal, M. J.
[6] R. H. Waterson, E. S. Lander, R. K. Wilson, Nature 2005, 437, 69.
[7] C. G. Elsik, R. L. Tellam, K. C. Worley, R. A. Gibbs, D. M. Muzny, G. M. McConnell, G. Sapiro, A. Aspuru-Guzik, J. T. Baker, M. Bastiani, J. D.
Weinstock, D. L. Adelson, E. E. Eichler, L. Elnitski, R. Guigó, D. L. Murray, S. N. Sotiropoulos, J. Taylor, G. Senthil, T. Lehner, M. B.
Hamernik, S. M. Kappes, H. A. Lewin, D. J. Lynn, F. W. Nicholas, A. Gerstein, A. W. Harrow, Nat. Methods 2021, 18, 701.
Reymond, M. Rijnkels, L. C. Skow, E. M. Zdobnov, L. Schook, J. Womack, [17] A. K. Fedorov, M. S. Gelfand, Nat. Comput. Sci. 2021, 1, 114.
T. Alioto, S. E. Antonarakis, A. Astashyn, C. E. Chapple, H.-C. Chen, J. [18] V. Marx, Nat. Methods 2021, 18, 715.
Chrast, F. Câmara, O. Ermolaeva, C. N. Henrichsen, W. Hlavina, Y. [19] N. Lambert, Y.-N. Chen, Y.-C. Cheng, G.-Y. Li, Che-Mingand Chen, F.
Kapustin, B. Kiryutin, P. Kitts, F. Kokocinski, M. Landrum, D. Maglott, K. Nori, Nat. Phys. 2013, 9, 10.
Pruitt, V. Sapojnikov, S. M. Searle, V. Solovyev, A. Souvorov, C. Ucla, C. [20] M. Mosheni, Y. Omar, G. S. Engel, M. B. Plenio, Quantum Effects in
Wyss, J. M. Anzola, D. Gerlach, E. Elhaik, D. Graur, J. T. Reese, R. C. Biology, Cambridge University Press 2014.
Edgar, J. C. McEwan, G. M. Payne, J. M. Raison, T. Junier, E. V. [21] A. Marais, B. Adams, A. K. Ringsmuth, M. Ferretti, J. M. Gruber, R.
Kriventseva, E. Eyras, M. Plass, R. Donthu, D. M. Larkin, J. Reecy, M. Q. Hendrikx, M. Schuld, S. L. Smith, I. Sinayskiy, T. P. J. Krüger, F.
Yang, L. Chen, Z. Cheng, C. G. Chitko-McKown, G. E. Liu, L. K. Petruccione, R. van Grondelle, J. R. Soc. Interface 2018, 15, 20180640.
Matukumalli, J. Song, B. Zhu, D. G. Bradley, F. S. Brinkman, L. P. Lau, [22] J. Cao, R. J. Cogdell, D. F. Coker, H.-G. Duan, J. Hauer, U. Kleinekathöfer,
M. D. Whiteside, A. Walker, T. T. Wheeler, T. Casey, J. B. German, D. G. T. L. C. Jansen, T. Mančal, R. J. D. Miller, J. P. Ogilvie, V. I. Prokhorenko,
Lemay, N. J. Maqbool, A. J. Molenaar, S. Seo, P. Stothard, C. L. Baldwin, T. Renger, H.-S. Tan, R. Tempelaar, M. Thorwart, E. Thyrhaug, S.
R. Baxter, C. L. Brinkmeyer-Langford, W. C. Brown, C. P. Childers, T. Westenhoff, D. Zigmantas, Sci. Adv. 2020, 6, eaaz4888.
Connelley, S. A. Ellis, K. Fritz, E. J. Glass, C. T. Herzig, A. Iivanainen, K. K. [23] T. Vreven, K. Morokuma, Chapter 3 Hybrid Methods: ONIOM (QM:MM)
Lahmers, A. K. Bennett, C. M. Dickens, J. G. Gilbert, D. E. Hagen, H. Salih, and QM/MM, Annu. Rep. Comput. Chem. 2006, 2, 35.
J. Aerts, A. R. Caetano, B. Dalrymple, J. F. Garcia, C. A. Gill, S. G. [24] H. M. Senn, W. Thiel, Angew. Chem. Int. Ed. 2009, 48, 1198.
Hiendleder, E. Memili, D. Spurlock, J. L. Williams, L. Alexander, M. J. [25] U. Ryde, Chapter Six – QM/MM Calculations on Proteins, in G. A. Voth
Brownstein, L. Guan, R. A. Holt, S. J. Jones, M. A. Marra, R. Moore, S. S. (Editor), Computational Approaches for Studying Enzyme Mechanism
Moore, A. Roberts, M. Taniguchi, R. C. Waterman, J. Chacko, M. M. Part A, volume 577 of Methods Enzymol., pages 119–158, Academic
Chandrabose, A. Cree, M. D. Dao, H. H. Dinh, R. A. Gabisi, S. Hines, J. Press 2016.
Hume, S. N. Jhangiani, V. Joshi, C. L. Kovar, L. R. Lewis, Y. shin Liu, J. [26] S. Izrailev, S. Stepaniants, B. Isralewitz, D. Kosztin, H. Lu, F. Molnar, W.
Lopez, M. B. Morgan, N. B. Nguyen, G. O. Okwuonu, S. J. Ruiz, J. Wriggers, K. Schulten, Steered Molecular Dynamics, in P. Deuflhard, J.
Santibanez, R. A. Wright, C. Buhay, Y. Ding, S. Dugan-Rocha, J. Hermans, B. Leimkuhler, A. E. Mark, S. Reich, R. D. Skeel (Editors),
Herdandez, M. Holder, A. Sabo, A. Egan, J. Goodell, K. Wilczek-Boney, Computational Molecular Dynamics: Challenges, Methods, Ideas, Spring-
G. R. Fowler, M. E. Hitchens, R. J. Lozado, C. Moen, D. Steffen, J. T. er Berlin Heidelberg, Berlin, Heidelberg 1999 pages 39–65.
Warren, J. Zhang, R. Chiu, J. E. Schein, K. J. Durbin, P. Havlak, H. Jiang, [27] B. Isralewitz, M. Gao, K. Schulten, Curr. Opin. Struct. Biol. 2001, 11, 224.
Y. Liu, X. Qin, Y. Ren, Y. Shen, H. Song, S. N. Bell, C. Davis, A. J. Johnson, [28] T. Hansson, C. Oostenbrink, W. F. van Gunsteren, Curr. Opin. Struct. Biol.
S. Lee, L. V. Nazareth, B. M. Patel, L.-L. Pu, S. Vattathil, R. L. Williams, S. 2002, 12, 190.

ChemBioChem 2023, 24, e202300120 (21 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

[29] J. E. Stone, D. J. Hardy, I. S. Ufimtsev, K. Schulten, J. Mol. Graphics [79] V. Abraham, N. J. Mayhall, J. Chem. Theory Comput. 2020, 16, 6098.
Modell. 2010, 29, 116. [80] B. Huron, J. P. Malrieu, P. Rancurel, J. Chem. Phys. 1973, 58, 5745.
[30] K. Meier, A. Choutko, J. Dolenc, A. P. Eichenberger, S. Riniker, W. F. [81] S. Szalay, M. Pfeffer, V. Murg, G. Barcza, F. Verstraete, R. Schneider, Ö.
van Gunsteren, Angew. Chem. Int. Ed. 2013, 52, 2820. Legeza, Int. J. Quantum Chem. 2015, 115, 1342.
[31] D. J. Tannor, Introduction to Quantum Mechanics - A Time-Dependent [82] G. K.-L. Chan, J. J. Dorando, D. Ghosh, J. Hachmann, E. Neuscamman, H.
Perspective, University Science Books, Sausalito, California 2007. Wang, T. Yanai, An Introduction to the Density Matrix Renormalization
[32] D. Sholl, J. A. Steckel, Density functional theory, Wiley-Blackwell, Group Ansatz in Quantum Chemistry, in Frontiers in Quantum Systems
Hoboken, NJ 2009. in Chemistry and Physics, pages 49–65, Springer-Verlag 2008.
[33] T. Helgaker, P. Jorgensen, J. Olsen, Molecular electronic-structure theory, [83] G. K. L. Chan, D. Zgid, Annu. Rep. Comput. Chem. 2009, 5, 149.
John Wiley & Sons, Nashville, TN 2013. [84] K. H. Marti, M. Reiher, Z. Phys. Chem. 2010, 224, 583.
[34] M. Reiher, A. Wolf, Relativistic Quantum Chemistry – The Fundamental [85] U. Schollwöck, Ann. Phys. 2011, 326, 96.
Theory of Molecular Science, John Wiley & Sons, New York 2015. [86] G. K.-L. Chan, S. Sharma, Annu. Rev. Phys. Chem. 2011, 62, 465.
[35] C. David Sherrill, H. F. Schaefer, The Configuration Interaction Method: [87] S. Wouters, D. Van Neck, Eur. Phys. J. D 2013, 31, 272.
Advances in Highly Correlated Approaches, in P.-O. Löwdin, J. R. Sabin, [88] Y. Kurashige, Mol. Phys. 2014, 112, 1485.
M. C. Zerner, E. Brändas (Editors), Advances in Quantum Chemistry, [89] R. Olivares-Amaya, W. Hu, N. Nakatani, S. Sharma, J. Yang, G. K.-L. Chan,
volume 34, pages 143–269, Academic Press 1999. J. Chem. Phys. 2015, 142, 34102.
[36] R. J. Bartlett, M. Musiał, Rev. Mod. Phys. 2007, 79, 291. [90] T. Yanai, Y. Kurashige, W. Mizukami, J. Chalupský, T. N. Lan, M. Saitow,
[37] D. Cremer, WIREs Comput. Mol. Sci. 2011, 1, 509. Int. J. Quantum Chem. 2015, 115, 283.
[38] A. J. Cohen, P. Mori-Sánchez, W. Yang, Science 2008, 321, 792. [91] S. R. White, Phys. Rev. Lett. 1992, 69, 2863.
[39] X. Ren, P. Rinke, C. Joas, M. Scheffler, J. Mater. Sci. 2012, 47, 7447. [92] S. R. White, Phys. Rev. B 1993, 48, 10345.
[40] M. Reiher, Isr. J. Chem. 2022, 62, e202100101. [93] A. Baiardi, C. J. Stein, V. Barone, M. Reiher, J. Chem. Theory Comput.
[41] C. Brunken, M. Reiher, J. Chem. Theory Comput. 2021, 17, 3797. 2017, 13, 3764.
[42] K.-S. Csizi, M. Reiher, WIREs Comput. Mol. Sci. 2023, page e1656. [94] S. Paeckel, T. Köhler, A. Swoboda, S. R. Manmana, U. Schollwöck, C.
[43] J. Tomasi, B. Mennucci, R. Cammi, Chem. Rev. 2005, 105, 2999. Hubig, Ann. Phys. 2019, 411, 167998.
[44] A. Klamt, WIREs Comput. Mol. Sci. 2011, 1, 699. [95] A. Baiardi, M. Reiher, J. Chem. Theory Comput. 2019, 15, 3481.
[45] F. R. Manby, M. Stella, J. D. Goodpaster, T. F. Miller, J. Chem. Theory [96] P. Pulay, T. P. Hamilton, J. Chem. Phys. 1988, 88, 4926.
Comput. 2012, 8, 2564. [97] V. Veryazov, P. Å. Malmqvist, B. O. Roos, Int. J. Quantum Chem. 2011,
[46] B. Mennucci, WIREs Comput. Mol. Sci. 2012, 2, 386. 111, 3329.
[47] C. R. Jacob, J. Neugebauer, WIREs Comput. Mol. Sci. 2014, 4, 325. [98] E. R. Sayfutyarova, Q. Sun, G. K.-L. Chan, G. Knizia, J. Chem. Theory
[48] A. Warshel, Angew. Chem. Int. Ed. 2014, 53, 10020. Comput. 2017, 13, 4063.
[49] T. A. Wesolowski, S. Shedge, X. Zhou, Chem. Rev. 2015, 115, 5891. [99] C. J. Stein, M. Reiher, J. Comput. Chem. 2019, 40, 2216.
[50] C. Curutchet, B. Mennucci, Chem. Rev. 2017, 117, 294. [100] C. A. Coulson, Rev. Mod. Phys. 1960, 32, 170.
[51] A. H. Mühlbach, M. Reiher, J. Chem. Phys. 2018, 149, 184104. [101] R. H. Tredgold, Phys. Rev. 1957, 105, 1421.
[52] L. O. Jones, M. A. Mosquera, G. C. Schatz, M. A. Ratner, J. Am. Chem. [102] A. J. Coleman, Rev. Mod. Phys. 1963, 35, 668.
Soc. 2020, 142, 3281. [103] Y.-K. Liu, M. Christandl, F. Verstraete, Phys. Rev. Lett. 2007, 98, 110503.
[53] D. Loco, L. Lagardère, O. Adjoua, J.-P. Piquemal, Acc. Chem. Res. 2021, [104] A. Coleman, V. Yukalov, Reduced Density Matrices: Coulson’s Challenge,
54, 2812. Lecture Notes in Chemistry, Springer 2000.
[54] T. P. Senftle, S. Hong, M. M. Islam, S. B. Kylasa, Y. Zheng, Y. K. Shin, C. [105] J. Cioslowski, Many-Electron Densities and Reduced Density Matrices,
Junkermeier, R. Engel-Herbert, M. J. Janik, H. M. Aktulga, T. Verstraelen, Mathematical and Computational Chemistry (MACC), Springer 2000.
A. Grama, A. C. T. van Duin, npj Comput. Mater. 2016, 2, 15011. [106] D. A. Mazziotti, Acc. Chem. Res. 2006, 39, 207.
[55] L. Bertini, M. Bruschi, L. de Gioia, P. Fantucci, C. Greco, G. Zampella, [107] T. Helgaker, S. Coriani, P. Jørgensen, K. Kristensen, J. Olsen, K. Ruud,
Quantum Chemical Investigations of Reaction Paths of Metalloenzymes Chem. Rev. 2012, 112, 543.
and BiomimeticModels – The Hydrogenase Example, pages 1–46, Spring- [108] R. P. Feynman, Int. J. Theor. Phys. 1982, 21, 467.
er Berlin Heidelberg, Berlin, Heidelberg 2007. [109] M. Fingerhuth, T. Babej, C. Ing, arXiv 2018, article no. 1810.13411.
[56] M. R. A. Blomberg, T. Borowski, F. Himo, R.-Z. Liao, P. E. M. Siegbahn, [110] A. Robert, P. K. Barkoutsos, S. Woerner, I. Tavernelli, npj Quantum Inf.
Chem. Rev. 2014, 114, 3601. 2021, 7, 38.
[57] M. Orio, D. A. Pantazis, Chem. Commun. 2021, 57, 3952. [111] S. Lloyd, Science 1996, 273, 1073.
[58] M. Born, V. Fock, Z. Phys. 1928, 51, 165. [112] T. S. Cubitt, A. Montanaro, S. Piddock, Proc. Natl. Acad. Sci. USA 2018,
[59] J. Schliemann, J. I. Cirac, M. Kuś, M. Lewenstein, D. Loss, Phys. Rev. A 115, 9497.
2001, 64, 022303. [113] I. Cirac, P. Zoller, Nat. Phys. 2012, 8, 264.
[60] M. A. Nielsen, I. L. Chuang, Quantum Computation and Quantum [114] A. Daley, I. Bloch, C. Kokail, N. Pearson, M. Troyer, P. Zoller, Nature
Information, Cambridge University Press, Cambridge, England 2010. 2022, 607, 667.
[61] M. Altunbulak, A. Klyachko, Commun. Math. Phys. 2008, 282. [115] P. W. Shor, Phys. Rev. A 1995, 52, R2493.
[62] K. Boguslawski, P. Tecmer, O. Legeza, M. Reiher, J. Phys. Chem. Lett. [116] D. Aharonov, M. Ben-Or, arXiv 1999, article no. 10.48550.
2012, 3, 3129. [117] O. Fawzi, A. Grospellier, A. Leverrier, Commun. ACM 2020, 64, 106–114.
[63] C. J. Stein, M. Reiher, J. Chem. Theory Comput. 2016, 12, 1760. [118] A. Y. Kitaev, Russ. Math. Surv. 1997, 52, 1191.
[64] C. Schilling, M. Altunbulak, S. Knecht, A. Lopes, J. D. Whitfield, M. [119] A. Kitaev, Ann. Phys. 2003, 303, 2.
Christandl, D. Gross, M. Reiher, Phys. Rev. A 2018, 97, 052503. [120] J. Preskill, Quantum 2018, page 79.
[65] A. Baiardi, M. Reiher, J. Chem. Phys. 2020, 152, 040903. [121] P. Jordan, E. Wigner, Zeitschrift für Physik 1928, 47, 631.
[66] J. W. Park, R. Al-Saadon, M. K. MacLeod, T. Shiozaki, B. Vlaisavljevich, [122] S. B. Bravyi, A. Y. Kitaev, Ann. Phys. 2002, 298, 210.
Chem. Rev. 2020, 120, 5878. [123] J. T. Seeley, M. J. Richard, P. J. Love, J. Chem. Phys. 2012, 137, 224109.
[67] A. Khedkar, M. Roemelt, Phys. Chem. Chem. Phys. 2021, 23, 17097. [124] J. R. McClean, J. Romero, R. Babbush, A. Aspuru-Guzik, New J. Phys.
[68] S. Grimme, WIREs Comput. Mol. Sci. 2011, 1, 211. 2016, 18, 023023.
[69] L. Kong, F. A. Bischoff, E. F. Valeev, Chem. Rev. 2012, 112, 75. [125] M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii,
[70] K. D. Vogiatzis, D. Ma, J. Olsen, L. Gagliardi, W. A. de Jong, J. Chem. J. R. McClean, K. Mitarai, X. Yuan, L. Cincio, P. J. Coles, Nat. Rev. Phys.
Phys. 2017, 147, 184111. 2021, 3, 625.
[71] I. Shavitt, Mol. Phys. 1998, 94, 3. [126] A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. Zhou, P. J. Love,
[72] R. J. Bartlett, M. Musiał, Rev. Mod. Phys. 2007, 79, 291. A. Aspuru-Guzik, J. L. O’Brien, Nat. Commun. 2014, 5, 4213.
[73] J. J. Eriksen, J. Phys. Chem. Lett. 2021, 12, 418. [127] A. Kandala, A. Mezzacapo, K. Temme, M. Takita, M. Brink, J. M. Chow,
[74] G. H. Booth, A. J. W. Thom, A. Alavi, J. Chem. Phys. 2009, 131, 054106. J. M. Gambetta, Nature 2017, 549, 242.
[75] F. Neese, J. Chem. Phys. 2003, 119, 9428. [128] J. Tilly, H. Chen, S. Cao, D. Picozzi, K. Setia, Y. Li, E. Grant, L. Wossnig, I.
[76] A. A. Holmes, N. M. Tubman, C. J. Umrigar, J. Chem. Theory Comput. Rungger, G. H. Booth, J. Tennyson, Phys. Rep. 2022, 986, 1.
2016, 12, 3674. [129] M. Benedetti, M. Fiorentini, M. Lubasch, Phys. Rev. Research 2021, 3,
[77] P. M. Zimmerman, J. Chem. Phys. 2017, 146, 104102. 033083.
[78] J. B. Schriber, F. A. Evangelista, J. Chem. Theory Comput. 2017, 13, 5354. [130] S. Barison, F. Vicentini, G. Carleo, Quantum 2021, 5, 512.

ChemBioChem 2023, 24, e202300120 (22 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

[131] S. McArdle, T. Jones, S. Endo, Y. Li, S. C. Benjamin, X. Yuan, npj [176] A. Potočnik, A. Bargerbos, F. A. Y. N. Schröder, S. A. Khan, M. C. Collodo,
Quantum Information 2019, 5, 75. S. Gasparinetti, Y. Salathé, C. Creatore, C. Eichler, H. E. Türeci, A. W.
[132] A. Aspuru-Guzik, A. D. Dutoi, P. J. Love, M. Head-Gordon, Science 2005, Chin, A. Wallraff, Nat. Commun. 2018, 9, 904.
309, 1704. [177] W. R. P. Scott, P. H. Hünenberger, I. G. Tironi, A. E. Mark, S. R. Billeter, J.
[133] H. Wang, S. Kais, A. Aspuru-Guzik, M. R. Hoffmann, Phys. Chem. Chem. Fennen, A. E. Torda, T. Huber, P. Krüger, W. F. van Gunsteren, J. Phys.
Phys. 2008, 10, 5388. Chem. A 1999, 103, 3596.
[134] L. Veis, J. Pittner, J. Chem. Phys. 2010, 133, 194106. [178] J. Wang, P. Cieplak, P. A. Kollman, J. Comput. Chem. 2000, 21, 1049.
[135] L. Veis, J. Višňák, T. Fleig, S. Knecht, T. Saue, L. Visscher, J. Pittner, Phys. [179] J. W. Ponder, D. A. Case, Force Fields for Protein Simulations, in Protein
Rev. A 2012, 85, 030304. Simulations, volume 66 of Adv. Protein Chem., pages 27–85, Academic
[136] L. Veis, J. Višňák, H. Nishizawa, H. Nakai, J. Pittner, Int. J. Quantum Press 2003.
Chem. 2016, 116, 1328. [180] P. S. Nerenberg, T. Head-Gordon, Curr. Opin. Struct. Biol. 2018, 49, 129.
[137] S. Lee, J. Lee, H. Zhai, Y. Tong, A. M. Dalzell, A. Kumar, P. Helms, J. Gray, [181] J. G. Kirkwood, J. Chem. Phys. 1935, 3, 300.
Z.-H. Cui, W. Liu, M. Kastoryano, R. Babbush, J. Preskill, D. R. Reichman, [182] M. Motta, C. Sun, A. T. K. Tan, M. J. O’Rourke, E. Ye, A. J. Minnich,
E. T. Campbell, E. F. Valeev, L. Lin, G. K.-L. Chan, arXiv 2022, article F. G. S. L. Brandão, G. K.-L. Chan, Nat. Phys. 2020, 16, 205.
no. 2208.02199. [183] K. Temme, T. J. Osborne, K. G. Vollbrecht, D. Poulin, F. Verstraete,
[138] G. H. Low, I. L. Chuang, Phys. Rev. Lett. 2017, 118, 010501. Nature 2011, 471, 87.
[139] A. M. Childs, Y. Su, Phys. Rev. Lett. 2019, 123, 050503. [184] D. Frenkel, B. Smit, Statistical Mechanics, in Understanding Molecular
[140] M. Reiher, N. Wiebe, K. M. Svore, D. Wecker, M. Troyer, Proc. Nat. Acad. Simulation, pages 9–22, Elsevier 2002.
Sci. 2017, 114, 7555. [185] C. Micheletti, P. Hauke, P. Faccioli, Phys. Rev. Lett. 2021, 127, 080501.
[141] K. Bourzac, Chem. Eng. News 2017, 95, 43. [186] D. Ghamari, P. Hauke, R. Covino, P. Faccioli, Sci. Rep. 2022, 12, 16336.
[142] V. von Burg, G. H. Low, T. Häner, D. S. Steiger, M. Reiher, M. Roetteler, [187] Y. Duan, P. A. Kollman, Science 1998, 282, 740.
M. Troyer, Phys. Rev. Research 2021, 3, 033055. [188] C. D. Snow, H. Nguyen, V. S. Pande, M. Gruebele, Nature 2002, 420,
[143] D. Rappoport, C. J. Galvin, D. Y. Zubarev, A. Aspuru-Guzik, J. Chem. 102.
Theory Comput. 2014, 10, 897. [189] K. Lindorff-Larsen, S. Piana, R. O. Dror, D. E. Shaw, Science 2011, 334,
[144] G. N. Simm, M. Reiher, J. Chem. Theory Comput. 2017, 13, 6108. 517.
[145] G. Wächtershäuser, Microbiol. Rev. 1988, 52, 452. [190] F. D. Malone, R. M. Parrish, A. R. Welden, T. Fox, M. Degroote, E.
[146] G. Wächtershäuser, Prog. Biophys. Mol. Biol. 1992, 58, 85. Kyoseva, N. Moll, R. Santagati, M. Streif, arXiv 2021, article no.
[147] V. Barone, M. Biczysko, C. Puzzarini, Acc. Chem. Res. 2015, 48, 1413. 2110.01589.
[148] K. I. Öberg, Chem. Rev. 2016, 116, 9631. [191] J. J. M. Kirsopp, C. Di Paola, D. Z. Manrique, M. Krompiec, G. Greene-
[149] S. A. Sandford, M. Nuevo, P. P. Bera, T. J. Lee, Chem. Rev. 2020, 120, Diniz, W. Guba, A. Meyder, D. Wolf, M. Strahm, D. Muñoz Ramo, Int. J.
4616. Quantum Chem. 2022, 122, e26975.
[150] C. F. Chyba, K. P. Hand, Ann. Rev. Astron. Astrophys. 2005, 43, 31. [192] A. S. Fraenkel, Bull. Math. Biol. 1993, 55, 1199.
[151] J. A. Pople, Rev. Mod. Phys. 1999, 71, 1267. [193] D. A. Hinds, M. Levitt, Proc. Natl. Acad. Sci. USA 1992, 89, 2536.
[152] W. Kohn, Rev. Mod. Phys. 1999, 71, 1253. [194] I. Halperin, B. Ma, H. Wolfson, R. Nussinov, Proteins Struct. Funct. Genet.
[153] C. E. Dykstra, G. Frenking, K. Kim, G. Scuseria (Editors), Theory and 2002, 47, 409.
applications of computational chemistry, Elsevier Science, London, [195] R. Dias, W. F. de Azevedo Jr, Curr. Drug Targets 2008, 9, 1040.
England 2005. [196] C. H. Bennett, E. Bernstein, G. Brassard, U. Vazirani, SIAM J. Comput.
[154] P. Horton, A. V. Ruban, R. G. Walters, Annu. Rev. Plant Physiol. 1996, 47, 1997, 26, 1510.
655. [197] L. K. Grover, Phys. Rev. Lett. 1997, 79, 325.
[155] Y.-C. Cheng, G. R. Fleming, Ann. Rev. Phys. Chem. 2009, 60, 241. [198] E. Farhi, J. Goldstone, S. Gutmann, arXiv 2014, article no. 1411.4028.
[156] G. D. Scholes, G. R. Fleming, A. Olaya-Castro, R. van Grondelle, Nat. [199] S. Lloyd, arXiv 2018, article no. 1812.11075.
Chem. 2011, 3, 763. [200] E. Farhi, A. W. Harrow, arXiv 2016, article no. 1602.07674.
[157] T. Ritz, S. Adem, K. Schulten, Biophys. J. 2000, 78, 707. [201] L. Zhou, S.-T. Wang, S. Choi, H. Pichler, M. D. Lukin, Phys. Rev. X 2020,
[158] S. Huelga, M. Plenio, Contemp. Phys. 2013, 54, 181. 10, 021067.
[159] C. Curutchet, B. Mennucci, Chem. Rev. 2017, 117, 294. [202] M. P. Harrigan, K. J. Sung, M. Neeley, K. J. Satzinger, F. Arute, K. Arya, J.
[160] T. Mirkovic, E. E. Ostroumov, J. M. Anna, R. van Grondelle, Govindjee, Atalaya, J. C. Bardin, R. Barends, S. Boixo, M. Broughton, B. B. Buckley,
G. D. Scholes, Chem. Rev. 2017, 117, 249. D. A. Buell, B. Burkett, N. Bushnell, Y. Chen, Z. Chen, B. Chiaro, R.
[161] G. S. Engel, T. R. Calhoun, E. L. Read, T.-K. Ahn, T. Mančal, Y.-C. Cheng, Collins, W. Courtney, S. Demura, A. Dunsworth, D. Eppens, A. Fowler, B.
R. E. Blankenship, G. R. Fleming, Nature 2007, 446, 782. Foxen, C. Gidney, M. Giustina, R. Graff, S. Habegger, A. Ho, S. Hong, T.
[162] G. Panitchayangkoon, D. Hayes, K. A. Fransted, J. R. Caram, E. Harel, J. Huang, L. B. Ioffe, S. V. Isakov, E. Jeffrey, Z. Jiang, C. Jones, D. Kafri, K.
Wen, R. E. Blankenship, G. S. Engel, Proc. Natl. Acad. Sci. USA 2010, 107, Kechedzhi, J. Kelly, S. Kim, P. V. Klimov, A. N. Korotkov, F. Kostritsa, D.
12766. Landhuis, P. Laptev, M. Lindmark, M. Leib, O. Martin, J. M. Martinis, J. R.
[163] E. Collini, C. Y. Wong, K. E. Wilk, P. M. G. Curmi, P. Brumer, G. D. Scholes, McClean, M. McEwen, A. Megrant, M. Mi, Xiaoand Mohseni, W.
Nature 2010, 463, 644. Mruczkiewicz, J. Mutus, O. Naaman, C. Neill, F. Neukart, M. Y. Niu, T. E.
[164] H.-G. Duan, V. I. Prokhorenko, R. J. Cogdell, K. Ashraf, A. L. Stevens, M. O’Brien, B. O’Gorman, E. Ostby, A. Petukhov, H. Putterman, C. Quintana,
Thorwart, R. J. D. Miller, Proc. Natl. Acad. Sci. USA 2017, 114, 8493. P. Roushan, N. C. Rubin, D. Sank, A. Skolik, V. Smelyanskiy, D. Strain, M.
[165] E. Thyrhaug, R. Tempelaar, M. J. P. Alcocer, K. Žídek, D. Bína, J. Streif, M. Szalay, A. Vainsencher, T. White, Z. J. Yao, P. Yeh, A. Zalcman,
Knoester, T. L. C. Jansen, D. Zigmantas, Nat. Chem. 2018, 10, 780. L. Zhou, H. Neven, D. Bacon, E. Lucero, E. Farhi, R. Babbush, Nat. Phys.
[166] A. W. Chin, J. Prior, R. Rosenbach, F. Caycedo-Soler, S. F. Huelga, M. B. 2021, 17, 332.
Plenio, Nat. Phys. 2013, 9, 113. [203] P. Hauke, H. G. Katzgraber, W. Lechner, H. Nishimori, W. D. Oliver, Rep.
[167] E. J. O’Reilly, A. Olaya-Castro, Nat. Commun. 2014, 5, 3012. Prog. Phys. 2020, 83, 054401.
[168] J. C. Dean, T. Mirkovic, Z. S. Toa, D. G. Oblinsky, G. D. Scholes, Chem [204] P. Hauke, G. Mattiotti, P. Faccioli, Phys. Rev. Lett. 2021, 126, 028104.
2016, 1, 858. [205] A. Perdomo-Ortiz, N. Dickson, M. Drew-Brook, G. Rose, A. Aspuru-Guzik,
[169] C. T. Rodgers, P. J. Hore, Proc. Nat. Acad. Sci. 2009, 106, 353. Sci. Rep. 2012, 2, 571.
[170] P. J. Hore, H. Mouritsen, Ann. Rev. Biophys. 2016, 45, 299. [206] R. Babbush, A. Perdomo-Ortiz, B. O’Gorman, W. Macready, A. Aspuru-
[171] T. P. Fay, L. P. Lindoy, D. E. Manolopoulos, P. J. Hore, Faraday Discuss. Guzik, Construction of Energy Functions for Lattice Heteropolymer
2020, 221, 77. Models: Efficient Encodings for Constraint Satisfaction Programming and
[172] J. Cai, M. B. Plenio, Phys. Rev. Lett. 2013, 111, 230503. Quantum Annealing, volume 155, chapter 5, pages 201–244, John
[173] A. M. Lewis, T. P. Fay, D. E. Manolopoulos, J. Chem. Phys. 2016, 145, Wiley & Sons, Ltd 2014.
244101. [207] P. A. M. Casares, R. Campos, M. A. Martin-Delgado, Quantum Sci.
[174] L. P. Lindoy, T. P. Fay, D. E. Manolopoulos, J. Chem. Phys. 2020, 152, Technol. 2022, 7, 025013.
164107. [208] R. Zwanzig, A. Szabo, B. Bagchi, Proc. Natl. Acad. Sci. USA 1992, 89, 20.
[175] B.-X. Wang, M.-J. Tao, Q. Ai, T. Xin, N. Lambert, D. Ruan, Y.-C. Cheng, F. [209] M. Rossmannek, P. K. Barkoutsos, P. J. Ollitrault, I. Tavernelli, J. Chem.
Nori, F.-G. Deng, G.-L. Long, npj Quantum Inf. 2018, 4, 52. Phys. 2021, 154, 114105.

ChemBioChem 2023, 24, e202300120 (23 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH
 Review
ChemBioChem doi.org/10.1002/cbic.202300120

[210] C. Vorwerk, N. Sheng, M. Govoni, B. Huang, G. Galli, Nat. Comput. Sci. [228] V. Havlíček, A. D. Córcoles, K. Temme, A. W. Harrow, A. Kandala, J. M.
2022, 2, 424. Chow, J. M. Gambetta, Nature 2019, 567, 209.
[211] M. Rossmannek, F. Pavošević, A. Rubio, I. Tavernelli, J. Phys. Chem. Lett. [229] Y. Liu, S. Arunachalam, K. Temme, Nat. Phys. 2021, 17, 1013.
2023, 14, 3491. [230] J. R. McClean, S. Boixo, V. N. Smelyanskiy, R. Babbush, H. Neven, Nat.
[212] Y. Liu, O. R. Meitei, Z. E. Chin, A. Dutt, M. Tao, T. V. Voorhis, I. L. Chuang, Commun. 2018, 9, 4812.
arXiv 2023, article no. 2301.01457. [231] C. Ortiz Marrero, M. Kieferová, N. Wiebe, PRX Quantum 2021, 2,
[213] J. Jumper, R. Evans, A. Pritzel, T. Green, M. Figurnov, O. Ronneberger, K. 040316.
Tunyasuvunakool, R. Bates, A. Žídek, A. Potapenko, A. Bridgland, C. [232] James M. Heather, Benjamin Chain, Genomics 2016, 107, 1.
Meyer, S. A. A. Kohl, A. J. Ballard, A. Cowie, B. Romera-Paredes, S. [233] E. W. Myers, Bioinformatics 2005, 21, ii79.
Nikolov, R. Jain, J. Adler, T. Back, S. Petersen, D. Reiman, E. Clancy, M. [234] P. E. C. Compeau, P. A. Pevzner, G. Tesler, Nat. Biotechnol. 2011, 29,
Zielinski, M. Steinegger, M. Pacholska, T. Berghammer, S. Bodenstein, 987.
D. Silver, O. Vinyals, A. W. Senior, K. Kavukcuoglu, P. Kohli, D. Hassabis, [235] N. Nagarajan, M. Pop, Nat. Rev. Genet. 2013, 14, 157.
Nature 2021, 596, 583. [236] D. MacLean, J. D. G. Jones, D. J. Studholme, Nat. Rev. Microbiol. 2009, 7.
[214] K. M. Ruff, R. V. Pappu, J. Mol. Biol. 2021, 433, 167208. [237] K. Prousalis, N. Konofaos, Sci. Rep. 2019, 9, 7226.
[215] C. J. Oldfield, A. K. Dunker, Annu. Rev. Biochem. 2014, 83, 553. [238] A. Sarkar, Z. Al-Ars, C. G. Almudever, K. Bertels, arXiv 2019, article
[216] R. B. Best, K. Lindorff-Larsen, M. A. DePristo, M. Vendruscolo, Proc. Nat. no. 1909.05563.
Acad. Sci. 2006, 103, 10901. [239] R. Babbush, J. R. McClean, M. Newman, C. Gidney, S. Boixo, H. Neven,
[217] T. Ching, D. S. Himmelstein, B. K. Beaulieu-Jones, A. A. Kalinin, B. T. Do, PRX Quantum 2021, 2, 010103.
G. P. Way, E. Ferrero, P.-M. Agapow, M. Zietz, M. M. Hoffman, W. Xie, [240] A. Sarkar, Z. Al-Ars, K. Bertels, PLoS One 2021, 16, 1.
G. L. Rosen, B. J. Lengerich, J. Israeli, J. Lanchantin, S. Woloszynek, A. E. [241] A. S. Boev, A. S. Rakitko, S. R. Usmanov, A. N. Kobzeva, I. V. Popov, V. V.
Carpenter, A. Shrikumar, J. Xu, E. M. Cofer, C. A. Lavender, S. C. Turaga, Ilinsky, E. O. Kiktenko, A. K. Fedorov, Sci. Rep. 2021, 11, 13183.
A. M. Alexandari, Z. Lu, D. J. Harris, D. DeCaprio, Y. Qi, A. Kundaje, Y. [242] R. Y. Li, R. Di Felice, R. Rohs, D. A. Lidar, npj Quantum Inf. 2018, 4, 14.
Peng, L. K. Wiley, M. H. S. Segler, S. M. Boca, S. J. Swamidass, A. Huang, [243] C. L. Degen, F. Reinhard, P. Cappellaro, Rev. Mod. Phys. 2017, 89,
A. Gitter, C. S. Greene, J. R. Soc. Interface 2018, 15, 20170387. 035002.
[218] J. G. Greener, S. M. Kandathil, L. Moffat, D. T. Jones, Nat. Rev. Mol. Cell [244] K. Jensen, R. Budvytyte, R. A. Thomas, T. Wang, A. M. Fuchs, M. V.
Biol. 2022, 23, 40. Balabas, G. Vasilakis, L. D. Mosgaard, H. C. Stærkind, J. H. Müller, T.
[219] J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, S. Lloyd, Heimburg, S.-P. Olesen, E. S. Polzik, Sci. Rep. 2016, 6, 29638.
Nature 2017, 549, 195. [245] D. Gottesman, T. Jennewein, S. Croke, Phys. Rev. Lett. 2012, 109,
[220] M. Cerezo, G. Verdon, H.-Y. Huang, L. Cincio, P. J. Coles, Nat. Comput. 070503.
Sci. 2022, 2, 567.
[221] S. Lloyd, M. Mohseni, P. Rebentrost, Nat. Phys. 2014, 10, 631.
[222] P. Rebentrost, M. Mohseni, S. Lloyd, Phys. Rev. Lett. 2014, 113, 130503.
[223] V. Giovannetti, S. Lloyd, L. Maccone, Phys. Rev. Lett. 2008, 100, 160501.
[224] S. Aaronson, Nat. Phys. 2015, 11, 291.
[225] I. Cong, S. Choi, M. D. Lukin, Nat. Phys. 2019, 15, 1273.
[226] N. Killoran, T. R. Bromley, J. M. Arrazola, M. Schuld, N. Quesada, S. Manuscript received: February 16, 2023
Lloyd, Phys. Rev. Research 2019, 1, 033063. Revised manuscript received: April 27, 2023
[227] A. Abbas, D. Sutter, C. Zoufal, A. Lucchi, A. Figalli, S. Woerner, Nat. Accepted manuscript online: May 7, 2023
Comput. Sci. 2021, 1, 403. Version of record online: June 1, 2023

ChemBioChem 2023, 24, e202300120 (24 of 24) © 2023 The Authors. ChemBioChem published by Wiley-VCH GmbH