Abstract Body

Machine learning and deep learning are part of a family of tools related to artificial intelligence and represent an emerging field of information and computer science that uses large data sets to extract features and representations. Antibody discovery and engineering is reliant on experimental platforms of high-throughput expression and screening of libraries. Here, I will describe how researchers are using machine and deep learning to assist in antibody engineering experiments and thus move beyond experimental screening. One area that I will highlight in particular is related to deep sequencing of natural antibody repertoires (derived from B cells of humans and mice), which has become a promising and powerful tool in basic immunology, immunodiagnostics, and the drug discovery process. However, identification of relevant information in these large datasets remains challenging. I will explain how deep learning is being used to identify patterns of antigen-specificity from antibody repertoires. Approaches such as unsupervised clustering and deep generative modeling are then being used to elucidate the antibody sequence space by generating thousands of novel and functional variants insilico. In addition deep learning is also being used to interrogate and predict antigen-specificity from a massive diversity of antibody sequence space from synthetic antibody repertoires (derived from surface display libraries). With its scalability and capacity to interrogate a vast protein sequence space, deep learning offers great potential for antibody discovery, engineering, and optimization.