Generative machine learning for designing antibody-based therapeutics
David Gifford and Michael Birnbaum lead an effort at MIT to develop novel machine learning methods to design human therapeutic antibodies based upon computationally designed experiments. These methods will propose antibodies for peptide-major histocompatibility complex (MHC) targets, enabling the rapid production of reagents that can be personalized for essentially any disease for a diverse array of monitoring and therapeutic applications. By combining recent discontinuities in machine learning capacity, sequencing based data collection, and the direct synthesis of candidate molecules, they will use computationally designed data collection at a new scale, enabling the creation of precisely specified training data for therapeutic design. Their general machine learning methods do not address a specific disease, but rather can be applied on a wide range of indications from cancer to infectious diseases to autoimmunity. These integrated components will enable an entirely new approach to therapeutic design based on innovative experiments designed to facilitate machine learning.