Abstract

The biotech industry has seen an explosion in the development of therapeutic antibodies in the last two decades, and today most of the best-selling drugs are monoclonal antibodies. The advantages of antibodies as therapeutics – namely their high affinity, specificity, potency, stability, manufacturability, and low toxicity – are compelling. Nevertheless, there are many fundamental challenges associated with co-optimizing their key properties due to strong inherent tradeoffs, such as affinity/specificity, affinity/stability, and species cross-reactivity tradeoffs, which greatly impede their development. I will discuss our progress in addressing these challenges in a rational and predictive manner using machine learning, including several examples of multi-objective optimization ranging from antibody drug development to efficient and long-lived brain delivery of IgGs using bispecific antibodies.