Bioinformatics, 39(2), 2023, btad083

https://doi.org/10.1093/bioinformatics/btad083
Advance Access Publication Date: 8 February 2023
Applications Note

Structural bioinformatics
APL@voro—interactive visualization and analysis of cell

Downloaded from https://academic.oup.com/bioinformatics/article/39/2/btad083/7031239 by B-On Consortium Portugal user on 11 April 2023

membrane simulations
Martin Kern1, Sabrina Jaeger-Honz1, Falk Schreiber 1,2,
* and Bjorn Sommer 3,
*
1
Department of Computer and Information Science, University of Konstanz, Konstanz 76484, Germany, 2Faculty of Information
Technology, Monash University, Clayton, VIC 3800, Australia and 3Royal College of Art, School of Design, London SW7 2EU, UK
*To whom correspondence should be addressed.
Associate Editor: Lenore Cowen
Received on May 31, 2022; revised on November 11, 2022; editorial decision on December 15, 2022; accepted on February 7, 2023

Abstract
Summary: Molecular dynamics (MD) simulations of cell membranes allow for a better understanding of complex
processes such as changing membrane dynamics, lipid rafts and the incorporation/passing of macromolecules into/
through membranes. To explore and understand cell membrane compositions, dynamics and processes, visual
analytics can help to interpret MD simulation data. APL@Voro is a software for the interactive visualization and
analysis of cell membrane simulations. Here, we present the new APL@Voro, which has been continuously devel-
oped since its initial release in 2013. We discuss newly implemented algorithms, methodologies and features, such
as the interactive comparison of related simulations and methods to assign lipids to either the upper or lower leaflet.
Availability and implementation: The current open-source version of APL@Voro can be downloaded from http://
aplvoro.com.
Contact: [email protected] or [email protected]

1 Introduction (Humphrey et al., 1996) or UCSF Chimera (Pettersen et al., 2004).

MEMBPlugin is a plugin for VMD that can calculate APL, MT,
Cell membranes are central biological structures that separate the in- order parameter, APL distribution and cholesterol tilt angle distribu-
terior of a cell from the outside, thereby protecting the cell from its tion. MEMBPlugin offers a user interface and includes the option to
environment (Campbell, 2002). They consist of a lipid bilayer and plot generated data. MemSurfer is a Python API that uses 3D point
contain various membrane proteins and small molecules. The prop- coordinates for Delaunay triangulations and surface parameteriza-
erties of a cell membrane can change over time, for example, during tion to represent membrane surfaces. This procedure gives the user
different stages of the development of the cell or due to changes in
direct access to the membrane surface to enable calculations, such as
its environment. Molecular dynamics (MD) simulations of cell mem-
MT and APL. The visualizations are however non-interactive which
branes are a common approach to understand membranes and
limits the exploration of the results. APL@Voro is developed with
related processes. This includes using MD simulations to understand
the goal of offering a software that unifies the analysis, interactive
and prioritize cell-penetrating peptides (Tran et al., 2021), cell mem-
visualization, exploration and comparison of multiple membrane
brane pore sealing (Zhang et al., 2019), membrane permeability
(Venable et al., 2019) and many more processes. simulations. Table 1 contains a comparison of the different tools
Software tools for the analysis of cell membrane simulations and their features.
play an important role when trying to understand the simulations.
Examples are GridMAT-MD (Allen et al., 2009), FATSLiM
(Buchoux, 2017), MEMBPlugin (Guixà-González et al., 2014),
2 Methods and implementation
MemSurfer (Bhatia et al., 2019) and APL@Voro (Lukat et al.,
2013). GridMAT-MD, like most tools, is a command-line tool that APL@Voro uses data obtained from GROMACS simulations
can calculate membrane thickness (MT) and area per lipid (APL) (Abraham et al., 2015) which are processed and then visualized for
using algorithms similar to those used in APL@Voro. FATSLiM, an- the analysis. It supports .PDB, .NDX, .XTC and .TRR formats.
other command-line tool, offers advanced algorithms to handle Other formats can be converted by using external tools such as
strongly curved membranes and vesicles. The output of both tools MDAnalysis (Gowers et al., 2016; Michaud-Agrawal et al., 2011).
can then be visualized as 2D graphs with external packages like The data processing can be separated into the following steps:
Xmgrace or Matplotlib (Hunter, 2007), but they do not provide dir-
ect internal visualization of membrane structures. To visualize the 1. leaflet detection,
membrane itself additional software is required, such as VMD 2. Delaunay triangulation for each leaflet,

C The Author(s) 2023. Published by Oxford University Press.

V 1
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unre-
stricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
2 M.Kern et al.

Table 1. A comparison of different tools and their features

Category Feature MDAnalysis FATSLiM GridMAT-MD MEMB-Plugin MEMSurfer APL@Voro

Membrane types Flat membranes      

Curved membranes      
Vesicles      
Comparisons Multiple membrane comp.      
Access GUI      
Command line      

Downloaded from https://academic.oup.com/bioinformatics/article/39/2/btad083/7031239 by B-On Consortium Portugal user on 11 April 2023

API      
Visualizations Voronoi      
Internal plotting      
Interactions Mouse interaction      
Linked views      
Interactive molecule selection      

tick: supported by the tool; cross: not supported by the tool.

Fig. 1. APL@Voro with two different membrane simulations. (1) Local simulation control panels which are used to open new views and synchronize simulations, (2) Voronoi
views visualizing the APL using the linear optimized color scale (LOCS) (Herman and Levkowitz, 1992) (maximum: White—1.6 nm2, minimum: Black—0.15 nm2) and the
blue cells represent user-selected areas (selected in 3., DPP), (3) selection panel with detailed information on each lipid, including area, local thickness and the key atom pos-
ition, (4) 2D plotting view showing the average APL of the outer leaflet over time for both simulations and (5) global simulation control

3. use of Delaunay triangulation to insert non-membrane atoms membrane proteins. The newly added position-based approach fits a
(e.g. protein) and calculate membrane thickness, surface to the membrane using polynomial regression and uses the
4. Voronoi diagram construction from Delaunay triangulation, position of key atoms relative to that surface to assign lipids to a
and leaflet. Its runtime complexity is Oðnd2 Þ, where n is the number of
lipid atoms, and d is the degree of the polynomial. This approach is
5. use of Voronoi diagram to calculate APL.
in practice a bit slower than the orientation-based method (around
The new version of APL@Voro is now capable of loading mul- 4%), but it proved more reliable.
tiple simulations to compare them side by side. Each imported simu- The trajectory needs to be loaded into computer memory to en-
lation appears in a list from where views can be opened. New views able responsive interactions, requiring a decent amount of memory
appear in a docking area that can be arranged by the user. Figure 1 space. Therefore, the new version allows the user to exclusively
shows an example of two imported simulations. Each simulation load relevant parts of a trajectory, as well as skip frames in order
has a Voronoi view showing the lower leaflet. The 2D view contains to manage memory usage. Loading 300 frames of a membrane
the average area per lipid of the outer leaflet over time for both sim- with 23 764 atoms takes an average of 3.5 s on a system running
ulations. APL@Voro can synchronize simulations along frames Ubuntu 20.04 with a AMD Ryzen 7 5800X CPU and 32 Gb of
which can be used to line up certain events (e.g. a substance entering 3200 MHz RAM.
the membrane) or account for different simulation lengths or time The Voronoi view has been overhauled. The legend has now a
steps. fixed size and will be displayed in the top left corner. The top right
APL@Voro originally detected leaflets only by estimating the corner is used to display additional information on the visualized
orientation of lipids (Lukat et al., 2013). This approach is imprecise leaflet. Also the grid rendering has been improved, the view can be
in some cases, e.g. for complex lipids such as lipid A or around dragged around by holding the middle mouse button. Hovering over
APL@voro 3

a Voronoi cell will display information on the associated lipid next comparison of multiple membrane simulations within one session,
to the pointer. The original color scale options (rainbow or none) better algorithms (e.g. for leaflet detection) and an updated intuitive
have been extended by three other color scale options. graphical user interface. APL@Voro therefore allows for easy
APL@Voro can also be used as a pure command-line tool. Due analysis of MD simulations and helps in exploring changing
to a lack of interactivity, there is no need to load trajectories into membrane dynamics, lipid rafts and membrane interactions of mac-
memory. Using the command-line functions will now only load one romolecules. It is a good basis for MD simulation analysis but also
frame for analysis at a time to save memory. allows for future developments such as exploring density profiles,
To summarize the previous paragraphs, APL@Voro 3.3.3 offers diffusion coefficients, membrane curvature and deuterium order
the following new features in comparison to the old APL@Voro
parameters.
version:

Downloaded from https://academic.oup.com/bioinformatics/article/39/2/btad083/7031239 by B-On Consortium Portugal user on 11 April 2023

• load multiple simulations for side-by-side comparison,
• frame synchronization, Funding
• position-based leaflet detection, This work was partly supported by the Deutsche Forschungsgemeinschaft
• overhauled Voronoi view: improved visualization, easier naviga- (DFG, German Research Foundation) [Project-ID 251654672—TRR 161].
tion, larger area,
• possibility to load only parts of the trajectory, Conflict of Interest: none declared.
• additional color scales (rainbow, LOCS, heated object and linear
grey),
References
• optimization of memory usage (i.e. code optimization and 3D
Abraham,M.J. et al. (2015) GROMACS: high performance molecular simula-
view removed) and
tions through multi-level parallelism from laptops to supercomputers.
• optimized memory usage (by limiting the amount of memory SoftwareX, 1, 19–25.
used during parsing) when using console. Allen,W. et al. (2009) GridMAT-MD: a grid-based membrane analysis tool
for use with molecular dynamics. J. Comput. Chem., 30, 1952–1958.
Bhatia,H. et al. (2019) MemSurfer: a tool for robust computation and charac-
3 Usage terization of curved membranes. J. Chem. Theory Comput., 15, 6411–6421.
APL@Voro calculates APL and MT and creates an interactive visu- Buchoux,S. (2017) FATSLiM: a fast and robust software to analyze MD simu-
alization of membrane simulations. The average APL is related to lations of membranes. Bioinformatics, 33, 133–134.
other membrane properties like acyl chain ordering, compressibility Campbell,N.A. (2002) Biology. The Benjamin-Cummings Publishing Company,
and molecular packaging (Patra et al., 2003). Changes in the average Reading, Massachusetts, USA.
APL and MT, which are strongly related (Moradi et al., 2019), can Gowers,R.J. et al. (2016) MDAnalysis: a Python package for the rapid analysis
be indicative of various processes taking place in the membrane, e.g. of molecular dynamics simulations. In: S. Benthall and S. Rostrup (ed.)
phase changes. Proceedings of the 15th Python Science Conference, Austin, Texas, USA,
APL@Voro visualizes membrane leaflets as a Voronoi diagram pp. 98–105.
in the so-called Voronoi views where APL, MT and lipid neighbor- Guixà-González,R. et al. (2014) MEMBPLUGIN: studying membrane com-
hood can be mapped to one of several color scales. Colors can also plexity in VMD. Bioinformatics, 30, 1478–1480.
be manually assigned to lipid types. The Voronoi view is comple- Herman,G.T. and Levkowitz,H. (1992) Color scales for image data. IEEE
mented by a table that contains detailed information on each lipid Comp. Graph. Appl., 12, 72–80.
as well as averages for the whole leaflet and the current selection Humphrey,W. et al. (1996) VMD – visual molecular dynamics. J. Mol.
Graphics, 14, 33–38.
(see Fig. 1(3)). The Voronoi view can reveal lipid rafts, protein ag-
Hunter,J.D. (2007) Matplotlib: a 2D graphics environment. Comput. Sci.
gregation, local anomalies, etc. The user can define a selection based
Eng., 9, 90–95.
on a combination of conditions such as lipid type, MT, APL and
Lukat,G. et al. (2013) APL@voro: a Voronoi-based membrane analysis tool
neighbors. This can be used to track lipids with certain properties,
for GROMACS trajectories. J. Chem. Inf. Model., 53, 2908–2925.
e.g. lipids that are in the gel phase. These selections are also used for
Michaud-Agrawal,N. et al. (2011) MDAnalysis: a toolkit for the analysis of
2D views where the user can plot the average membrane thickness
molecular dynamics simulations. J. Comput. Chem., 32, 2319–2327.
and area per lipid for all open simulations over time (Fig. 1(4)).
Moradi,S. et al. (2019) Shedding light on the structural properties of lipid
The results can be exported in various ways for further analysis:
bilayers using molecular dynamics simulation: a review study. RSC Adv., 9,
Voronoi views and 2D views as image file, simulation frames as ei- 4644–4658.
ther .TXT or .XML file, and 2D plots as .XVG file. Patra,M. et al. (2003) Molecular dynamics simulations of lipid bilayers:
APL@Voro is available as executable versions for Linux as well major artifacts due to truncating electrostatic interactions. Biophys. J., 84,
as Windows. For Mac OS X (including M1 chipset), Windows ver- 3636–3645.
sions can be used via CrossOver (https://www.codeweavers.com/). Pettersen,E. et al. (2004) UCSF chimera–a visualization system for exploratory
In the future, we are planning to provide native Mac OS X versions research and analysis. J. Comput. Chem., 25, 1605–1612.
as well. Tran,D. et al. (2021) Using molecular dynamics simulations to prioritize and
understand AI-generated cell penetrating peptides. Sci. Rep., 11, 10630.
Venable,R. et al. (2019) Molecular dynamics simulations of membrane perme-
4 Discussion and future development
ability. Chem. Rev., 119, 5954–5997.
APL@Voro provides analysis and interactive visualization of cell Zhang,L. et al. (2019) Molecular dynamics simulation of cell membrane pore
membrane simulations. The new developments enable direct sealing. Extreme Mech. Lett, 27, 83–93.