Each atom on the interface is regarded as a node in the graph

Each atom on the interface is regarded as a node in the graph. learning, Computational models, Antibody generation Increasing the binding affinity of an antibody to its target antigen is key for antibody therapeutics. Here the authors report a pretrainable geometric graph neural network, GearBind, and explore its potential in in silico antibody affinity maturation. == Introduction == Antibody plays a crucial role in the human immune system and serves as a powerful diagnostic and therapeutic tool, due to its ability to bind selectively and specifically to target antigens with high affinity. In vivo, antibodies go through affinity maturation, where the target-binding affinity gradually increases as a result of somatic hypermutation and clonal selection1. When a new antigen surfaces, therapeutic antibody leads repurposed from known antibodies or screened from a natural or de novo designed library often require in vitro affinity maturation to enhance their binding affinity to a desired, usually Tandutinib (MLN518) sub-nanomolar, level. Wet lab experimental methods for in vitro antibody affinity maturation usually involve constructing mutant libraries and screening with display technology25. These methods, while significantly improved during the past few years, are still labor-intensive and costly in general, taking 23 months or more to complete the process. Lets consider the combinatorial search space of possible mutations. There are usually 5060 residues on the complementarity-determining region (CDR) of an antibody, which are hypervariable in vivo and contribute to the majority of the binding free energyGbind6. Previous works show that multiple point mutations are often needed for successful Rabbit polyclonal to TrkB affinity maturation7,8. Performing experiments on all combinations of over a thousand possible point mutations in antibody CDR regions (60 residues 19 residues per residue) is difficult if not prohibitive. Therefore, a fast and accurate computational method for narrowing down the search space is much desired. Nevertheless, it is nontrivial for computational affinity maturation methods to balance speed and accuracy. Molecular dynamics methods based on empirical Tandutinib (MLN518) force fields912rely on human Tandutinib (MLN518) knowledge and abstractions to evaluate binding free energy changes after mutations. However, accurate models are often too slow to be used for ranking thousands of mutations (let alone their combinations). In recent years, machine learning, and particularly deep learning, has been demonstrated as a powerful tool capable of tackling this dilemma. Many machine-learning methods1318formulate the affinity maturation problem as a structure-based binding free energy change (, wherewtis short forwild typeandmtdenotesmutant) prediction problem. However, despite the importance of protein side-chain conformation to proteinprotein interaction, most existing methods model atom-level geometric information indirectly or incompletely, e.g., using hand-crafted features or residue-level features. These approaches inadequately address the intricate interplay between side-chain atoms. Another critical problem is the massive amount of paired binding affinity data required by machine-learning models for them to become accurate and reliable. To the best of our knowledge, the largest publicly available proteinprotein binding free energy change dataset, Structural Kinetic and Energetic database of Mutant Protein Interactions (SKEMPI) v2.019, contains only 7085 Gbindmeasurements on 348 protein complexes, a tiny amount compared to the training set sizes of foundational protein models, such as AlphaFold220and ESM221. To tackle the aforementioned challenges, we introduce GearBind, a pretrainable deep neural network that leverages multi-level geometric message passing to model the nuanced proteinprotein interactions. We utilize contrastive pretraining techniques on large-scale protein structural dataset to incorporate vital structural insights into the model (Fig.1). In silico experiments on SKEMPI and an independent test set demonstrate the superior performance of GearBind and the benefit of pretraining. We combine the GearBind models with previous state-of-the-art methods to create an ensemble model that achieves state-of-the-art performance on all metrics. The ablation study confirms the importance of key design choices within GearBind and the key role it played in the ensemble. We then use the GearBind-based ensemble to perform in silico affinity maturation for two antibodies with distinct formats and target antigens. Binding of the antibody CR3022 against the spike (S) protein of the Omicron SARS-CoV-2 variant is.