Pretrainable Geometric Graph Neural Network for Antibody Affinity Maturation

Abstract

In the realm of antibody therapeutics development, increasing the binding affinity of an antibody to its target antigen is a crucial task. This paper presents GearBind, a pretrainable deep neural network designed to be effective for in silico affinity maturation. Leveraging multi-level geometric message passing alongside contrastive pretraining on protein structural data, GearBind capably models the complex interplay of atom-level interactions within protein complexes, surpassing previous state-of-the-art approaches on SKEMPI v2 in terms of Pearson correlation, mean absolute error (MAE) and root mean square error (RMSE). In silico experiments elucidate that pretraining helps GearBind become sensitive to mutation-induced binding affinity changes and reflective of amino acid substitution tendency. Using an ensemble model based on pretrained GearBind, we successfully optimize the affinity of CR3022 to the spike (S) protein of the SARS-CoV-2 Omicron strain. Our strategy yields a high success rate with up to 17-fold affinity increase. GearBind proves to be an effective tool in narrowing the search space for in vitro antibody affinity maturation, underscoring the utility of geometric deep learning and adept pre-training in macromolecule interaction modeling.

GeoAB: Towards Realistic Antibody Design and Reliable Affinity Maturation

“…However, the structures of mutant types are usually unknown, RDE-PPI (Luo et al, 2023) avoids the problem and uses a structure-aware pretrained network on side-chain rotamers to assist the prediction. Besides, there are also works on designing pretraining tasks for ΔΔ G predictions (Hsu et al, 2022; Yang et al, 2023; Cai et al, 2023). In comparison, G EO AB also considers mutation effects attributing to structural flexibility but uses joint training strategies instead of pretraining paradigms.…”

“…To be specific, increasing works are emerging for antibody design, which considers antigens and conserved regions (non-CDRs) as context information, to achieve the co-design of the sequence and structure of the target CDRs (Kong et al, 2023a; Jin et al, 2021). Besides, based on the co-design models, the optimization of the antibody by mutating amino acids in the CDRs to enhance the binding affinity can be realized, which is called affinity maturation (Cai et al, 2023). Iterative Target Augmentation (ITA) is proposed to fulfill the affinity maturation tasks (Yang et al, 2020), by iteratively adding co-design models’ prediction of the mutant antibodies structures and sequences with higher affinity predicted by another pretrained model to the training set and thus to guide and retrain the co-design models to generate antibodies of high affinity.…”