REPTOR: accelerating antibody discovery and improving hits with machine learning

Antibody therapeutics are becoming increasingly important across a broad range of indications, yet their development requires discovery from a variety of difficult sources. Traditional technologies are over four decades old, while newer single-cell approaches for mining survivors are gaining traction in the wake of the SARS-Cov2 pandemic. However, all mainstream discovery approaches significantly limit the sampling of the in-vivo antibody immune response, thereby potentially missing important therapeutic candidates. Approaches to better deconvolute the antibody response with high-throughput sequencing technologies have begun to be applied for research uses. However, using these large-scale data to directly perform antibody discovery has remained elusive. We aim to develop software to streamline the incorporation of high-throughput sequencing into the three mainstream discovery approaches, thereby reducing time and increasing discovery success rate. These software-enabled enhancements will cover high-throughput sequencing for hybridoma discovery, enhanced enrichment analysis for display methods, and simplified workflow analysis for popular single-cell methods. The same type of repertoire sequencing can then be used in a different context to improve candidate antibodies by leveraging the natural improvements the host individual's immune system has already discovered. This expansion of existing candidates is enabled by the deep sequencing of antibody repertoires using next-generation sequencing technology that provides a window into the natural antibody evolution and optimization. These newly deep repertoires are able to be exploited by novel algorithms for analyzing the large antibody families produced, as well as advances in deep learning that enable large amounts of unlabeled data to be synthesized and used for model training to search both across antibody families for similarities, as well as within those families.