SORS: AI and simulation for antibody drug discovery

Following in the footsteps of computer vision and natural language processing, the use of deep learning in the fields of protein science and protein engineering is rapidly increasing. Antibodies, in particular, are not only crucial research subjects in drug discovery, vaccines, and immunology but also serve as excellent benchmarks for the effective application of deep learning in protein science due to the availability of big data on sequences, structures, functions, and physical properties. Recently, in antibody drug discovery, AI and simulation have become essential for improving the efficiency of protein engineering, designing lead antibodies, and assessing and improving developability and immunogenicity. This presentation will provide an overview of these applications, incorporating our recent research to outline the current landscape, challenges, and future perspectives. This includes developing a methodology to assess "human-likeness" of antibody structures from two-dimensional images using a Variational Autoencoder as an anomaly detection problem, suggesting the potential to assess protein stability through ultra-short molecular dynamics simulations, and examples where point mutations have successfully stabilized proteins based on changes in free energy due to amino acid mutations.