ESMFold2: New Model Revolutionizes Protein Structure Prediction and Therapeutic Design

ESMFold2 has been announced as a state-of-the-art structure prediction model that excels in predicting protein structures from single sequences and multiple sequence alignments (MSAs). The model demonstrates improved performance on benchmarks related to protein-protein interactions, particularly in predicting antibody-antigen complexes. It has been developed to facilitate protein biology research, achieving high success rates in designing miniprotein binders and single chain antibodies across five therapeutic targets relevant to cancer and immunology. Additionally, an extensive atlas of 6.8 billion proteins and 1.1 billion predicted structures will be released, enhancing the understanding of protein interactions crucial for drug discovery. The model utilizes advanced language modeling techniques to derive insights into protein biology without prior knowledge, highlighting its potential to significantly impact therapeutic design and accelerate scientific research.