Tamarind Bio

Read our guide on how to use Tamarind Bio to computationally design de novo antibodies and nanobodies! Longer text: https://lnkd.in/gSt6KD6a Our de novo design tool: https://lnkd.in/gmNy3Rm2 Original paper from IPD and Baker Lab that we deploy: https://lnkd.in/gqCY3Rin Highlights: The authors demonstrate computationally designed antibody and nanobody binders validated via cryo-EM and other biophysical methods, starting from a framework structure, a target structure, and a user-defined epitope. High throughput experimental screening remains a bottleneck, as the protocol is able to generate a large number of designed VHHs and scFvs, but the ability to identify which are binders (let alone high affinity ones), remains limited. The primary constraint here seems to be an ability to predicted the structure of the designed binder against the target. I suspect Chai-1 with restraints or AF3 may yield some improvements here! The whole protocol is effectively an antibody specific variant of the original RFdiffusion de novo minibinder design pipeline, so much of the tricks added on top of that methodology should be applicable here as well. Get in touch for easy to use, highly scalable[especially important for this one :)], protein/antibody design software. My calendar is on my Linkedin profile!

Commercially Available AI De Novo Antibody/Nanobody Design, Try it now! Work from the Baker Lab (IPD) is finally published and available for commercial use. The authors computationally generate nanomolar affinity antibody/scFv and nanobody/VHH binders for a given target and epitope, using specialized versions of RFdiffusion, ProteinMPNN, and RoseTTAFold. Cryo-EM and other biophysical methods confirmed that these designs not only folded correctly into Ig-like structures but also adopted the intended binding poses with exact CDR loop conformations. While the first rounds of designs had modest binding affinity, with some optimization, they reached some very promising results. Available on Tamarind Bio now: https://lnkd.in/gMFKhXiX Preprint: https://lnkd.in/gqCY3Rin Congratulations to Nathaniel Bennett, Joe Watson, Robert Ragotte and the whole team of authors. Get in touch to learn more!

Using Computational Methods to Solubilize Membrane Proteins Membrane proteins are not members of the soluble proteome, i.e. their structure cannot be maintained outside of the membrane/in solution. Traditionally, the solubilization would be done using detergent, but this has its own problems as well. As an alternative, below are two papers that employ AI to solubilize membrane proteins to maintain their structural features for further study. And how to use them on Tamarind Bio! I previously discussed how the AI tool, ProteinMPNN, can be used to generate de novo sequences that fold into a given input structure. A variant of this method, trained only on soluble proteins, called Soluble ProteinMPNN can be used to generate sequences that maintain the original structure, while also improving solubility. Computational design of soluble and functional membrane protein analogues The main focus of this approach is to generate alternative sequences to an input structure, that do not lose their structure in solution. The authors achieve designs for "complex protein topologies and [enrich] them with functionalities from membrane proteins, with high experimental success rates, leading to a de facto expansion of the functional soluble fold space" The authors use AlphaFold2 to continuously test different replacement sequences and qualitatively evaluate if the sequence's predicted structure matches that of the native structure. Then, the authors feed these predicted structures to Soluble ProteinMPNN and find that the sequences produced via MPNN show significant experimental success. See below on how to use this methodology on Tamarind: AlphaFold inversion step: https://lnkd.in/gCW3QuZJ Soluble ProteinMPNN: https://lnkd.in/gi99zbwi Solubilization of Membrane Proteins using designed protein WRAPS Another approach to this problem is to design proteins that replicate the effect of detergent, i.e. to keep the protein stable and water soluble in the absence of the cells's lipid bilayer. The authors introduce the de novo protein category "Water-soluble RFdiffused Amphipathic Proteins". To create these WRAPS, the authors first create idealized helical and beta-barrel backbones manually, that match the shape and size of the transmembrane domain of the target. Then they feed this to the RFdiffusion fold-conditioning protocol, which will use this structure and as a template and generate a biophysically feasible alternative to this idealized backbone. Then, the authors use RFdiffusion's partial diffusion protocol to diversify the generated structures to ensure complementarity to the starting structure. Since RFdiffusion produces structures without their corresponding residues, they use Soluble ProteinMPNN to generate a viable sequence to fold into this backbone. Finally, they use AlphaFold to evaluate the fitness of each design. Use this protocol on Tamarind: https://lnkd.in/e9gj2Rtt ------------ See the comments for citations!

We can computationally generate nanomolar de novo minibinders, now what? AI Proteins just announced their $400 million research collaboration with BMS, funding work to design de novo miniproteins for clinically valuable targets. I previously discussed the advantages and limitations of miniproteins as a modality (https://lnkd.in/eVYuxV4f), especially in terms of target diversity, immunogenicity, and stability. With this, it's still useful to consider what therapeutic applications minibinders might have. Below are some results for miniproteins being used with therapeutic applications. One somewhat straightforward use case is: "De novo design of picomolar SARS-CoV-2 miniprotein inhibitors". The Institute for Protein Design, University of Washington's vaccine, SKYCovione, is the world's first computationally designed protein medicine. High resistance rates due to antigen escape remain a challenge for Chimeric antigen receptor T cell (CAR-T) therapies. Mergen et. al. show an AI-designed de novo binder for CAR-T, targeting EGFR. The authors confirm activity in effector readouts, along showing success in targeting a mutated BCMA protein variant resistant to the clinically used bispecific antibody. The mini protein binder designed using RFdiffusion(https://lnkd.in/e9gj2Rtt) matches the functional properties of an scFv in cytotoxicity, cytokine secretion, and lysis of primary patient-derived cancer organoids. See our protein design tool(https://lnkd.in/e9gj2Rtt) for the metrics the authors used to rank the generated sequences. Minibinders' short half-lives and high stability make them attractive as a method to distribute different other therapeutics. For a small molecule attached to a miniprotein, we get the specificy of a biologic and the tissue penetration and clearance of a small molecule. Since minibinders are filtered through the liver very quickly, they also serve as useful radiopharmaceuticals. Aktis Oncology's AKY-1189 delivers Actinium-225 Nectin-4 expressing tumors. Aktis is in the clinical stage now! "[AI Proteins] engineer a 5.6 kDa miniprotein that binds selectively to CD123 [a soluble cytokine important for the immune system] with KD = 0.5 pM. [The miniprotein] is thermostable, able to be produced recombinantly from E. coli, and conjugated to a fluorescein molecule (FITC) via an engineered lysine residue. The miniprotein-directed CAR exhibited multiple advantages over both the antibody and the direct CAR-T, including increased efficacy with lower dose required of both the adapter and the CAR-T cells, improved tumor occupation, and less CAR-T exhaustion." ------ At Tamarind Bio, we provide automated de novo protein design software, running at massive throughput. Including many of the successful results above. Get in touch to learn more! 

Check out our blogpost and programmatic API for Boltz-1

Optimize your proteins' fitness by emulating evolution using protein language models!

The entire high-throughput hit identification process done for you by our bioinformaticians!

AFsample2 on Tamarind: https://lnkd.in/gJFnwNRY

Optimize your enzymes or protein therapeutics for binding affinity to a ligand of your choice, or design a de novo binder to your target!