Protein structure-based antibody and small molecule drug design

Abstract

Structure-based drug design is the design of proteins or small molecules that induce desired interaction with a target protein based on the structural information of protein. In this study propose two computational methods that designs an antibody and a small molecule as a drug for a given target protein structure. The first method, AbFlex, is an antibody design model for a given antigen structure. Antibodies are proteins that the immune system produces in response to foreign pathogens. Designing antibodies that specifically bind to antigens is a key step in developing antibody therapeutics. The complementarity determining regions (CDRs) of the antibody are mainly responsible for binding to the target antigen, and therefore must be designed to recognize the antigen. AbFlex can design CDR structures and sequences for antigen structures, and its prediction accuracy is better than other similar methods. Moreover, more than 38% of newly designed antibodies are estimated to have better binding energies for their antigens than wild types. The effectiveness of the model is attributed to two different strategies that are developed to overcome the difficulty associated with the scarcity of antibody-antigen complex structure data. One strategy is to use an equivariant graph neural network model that is more dataefficient. More importantly, a new data augmentation strategy based on the flexible definition of CDRs significantly increases the performance of the CDR prediction model. The second method to be introduced is called MORLD (Molecule Optimization by Reinforcement Learning and Docking) that automatically generates and optimizes molecules by combining reinforcement learning and docking to develop predicted novel inhibitors for a target protein. MORLD requires only a target protein structure and directly modifies (or generates) ligand structures to obtain higher predicted binding affinity for the target protein without any other training data. Using MORLD, we were able to generate novel potential inhibitors against discoidin domain receptor 1 kinase in less than two days on a moderate computer. We also demonstrated MORLD’s ability to generate novel potential agonists for the D4 dopamine receptor from scratch without virtual screening on an ultra-large compound library. In addition, we introduces the model, MORLD with CReM, with strengths in denovo generation of molecule. The free web server is available at https://morld.kaist.ac.kr.