Over the past three decades, synthetic antibody repertoires have advanced significantly, aiding the development of novel therapeutic and diagnostic antibodies. Currently, at least six antibodies derived from such repertoires are in clinical use, with hundreds more in clinical trials. Most repertoires contain billions of unique antibodies, generated by recombining fragments of known human antibodies. However, these repertoires often yield antibodies with low stability and a high tendency for polyreactivity (tendency to bind to unwanted targets), complicating the costly development process.
In their latest project, scientists from Sarel Fleishman's group are leveraging AI models to optimize various stages of synthetic antibody repertoire development. Their aim is to identify and learn to produce more stable, less polyreactive antibodies. The project's goal is to utilize AI's capabilities to enhance our understanding of the biophysical foundations of antibody structure and function, develop new methods for ranking drug candidates, and generate new and improved repertoires comprising billions of diverse human antibodies. This, in turn, can expedite and reduce the costs of discovering life-saving therapeutics.