MIT Develops PottsMPNN for Protein Design

MIT's Department of Biology has developed PottsMPNN, a machine-learning framework that advances protein design by focusing on the sequence-energy landscape. Unlike traditional methods that rely on native sequences, PottsMPNN generates novel protein structures by incorporating physical principles and evolutionary data. This approach enhances the prediction of protein stability and mutation effects, paving the way for designing proteins with no natural counterparts. The framework represents a significant step forward in using AI for biological engineering, potentially enabling the creation of entirely new proteins for various applications.