Scientists at Stanford University have developed a machine learning-based method that can predict molecular changes for the development of better antibody drugs. By combining the 3D structure of protein with large language models based on amino acid sequence, the researchers were able to quickly identify rare and desirable mutations that would otherwise require extensive experiments. The method was successfully applied to improve an FDA-approved SARS-CoV-2 antibody, resulting in a 25-fold increase in efficacy against the virus.

The approach utilizes structure, in addition to sequence, to guide the prediction algorithm and has the potential to accelerate the discovery of more effective medicines. The researchers have made their model and code freely available for others to use.