Free software: AlphaFold (structural protein database; predicts a proteins 3D structure from its amino acid sequence) Openfold3 (collaboration with Nvidia)
Latent-X (by Latent Lab is a de novo protein design model) SandboxAQ (announced the Structurally Augmented IC50 Repository (SAIR), an open access repository that leverages the Boltz seris of models to generate computationally folded protein ligand structures linke to corresponding experimental drug affinity values). Boltz-2 (predicts molecular binding affinity-Youtube video) RFdiffusion (enables users to create completely novel proteins based on molecualr specifications)
Paid software: ProteinMPNN (can take a protein structure and generate an amino acid sequence that will fold into that structure) ThermoMPNN (can predict change in protein stability due to point mutations).
Companies: Tamarind Bio (has built a web interfact where resarchers can access hudreds of AI tools, some open sourced and some licenced and get guidance and support in how to use them). Openfold (non-profit AI research and development consortium developing free and open-source software tools for biology and drug discovery.)
Introduction/Definitions:
computational docking: is based on maximizing the shape and chemical complementarities between a given pair of interacting proteins.
Artificial Intelligence:
Deep Cure (uses a plaform called MolGen which builds custom libraries for a specific set of requirements)
Typically scientists start by screening massive libaries of small molecuels with a hope of finding a starting point for a long optimization process. However, many companies are starting to use intentional enginenring such as wehn developing small molecule drugs for therapeutic targets. Instead of jsut screening massive libaries and hoping that something will stick, the drug is designed form the beginning to meet specific requirements.
Iktos generates innovative molecules by using an AI driven retrosyntehsis platform. The company uses generative AI to find ways of breaking down a complicated target molecule. Its proprietary generative AI, trained on millions of organic reacitons, generates molecules like a chemist by leveraging commercial building blocks and organic reactions over several steps.
Exscientia milestones include final results for Phase I/II trails for its cyclin-dependent kinase 7 inhibitor latter in 2024. The company beleives that AI driven drug design will result in unprecedented drugs as opposed to incremental gains in efficiency and speed.
De novo or ab initio methods:
De novo methods predicts the structure from sequence alone, without relying on similarity at the fold level between the modeled sequence and any known structures. These methods assume that the native structure corresponds to the global free-energy minimum and attempts to find this minimum by an exploration of many conceivable protein conformations.
Grinter discloses a RFdiffusion-based protein design to create binders that block hemoglobin binding to ChuA. design de novo protein binders to block heme acquisition from hemoglobin. Using an AlphaFold2 model of ChuA as a target, they utilised RFdiffusion and ProteinMPNN to design binders targeting extracellular loops 7 and 8 of ChuA, which accordingl to their model indicated were responsible for hemoglobin binding. They screened a limited number of these designs, identifying several binders that inhibit E. coli growth at low nanomolar concentrations when hemoglobin or myoglobin was the sole available iron source. See Grinter
Designing Antibody-Antigen Binders:
Using the 3-D structure of the antibody-antigen complexes, it is possible to enhance the antibody-antigen binding affinities by in silico mutations on antibody residues. In the best situation, when the antibody-antigen complex structures are available, it is relatively straight forward to perform affinity maturation in silico. First, the protein backbone is treated as rigid, and the conformation of the side chain was determined by discrete side-chain rotamer search. Second the lowest energy of the structures was further re-evaluated by using more accurate, but computationally more expense models. (Zhao, “In silico methods in antibody design” Antibodies, 2018)
Homology modeling:
Homology modeling relies on detectable similarity spanning most of the modeled sequence and at least one known structure. It relies on finding known structures related to the sequence to be modeled, aligning the sequence with the related structures, building a model, and assessing the model.
Bioinformatic analysis of genomic sequences:
Multiple sequence alignments and protein structure
–ConPLex: is an in silico screening tool which makes predictions of binding based on the distance between learned representations, enabling predictions at the scale of massive compound libraries and the human proteome. (Singh, Biophysics and computational Biology, “Constrastiv learning in protein language space predicts interactions between drugs and protein targets”, 120(24), 2023).