Key points are not available for this paper at this time.
Abstract AlphaFold3 has set the new state-of-the-art in predicting protein-protein complex structures. However, the complete picture of biomolecular interactions cannot be fully captured by static structures alone. In the field of protein engineering and antibody discovery, the connection from structure to function is often mediated by binding energy. This work benchmarks AlphaFold3 against SKEMPI, a commonly used binding energy dataset. We demonstrate that AlphaFold3 learns unique information and synergizes with force field, profile-based, and other deep learning methods in predicting the mutational effects on protein-protein interactions. We hypothesize that AlphaFold3 captures a more global effect of mutations by learning a smoother energy landscape, but it lacks the modeling of full atomic details that are better addressed by force field methods, which possess a more rugged energy landscape. Integrating both approaches could be an interesting future direction. All of our benchmark results are openly available at https: //github. com/luwei0917/AlphaFold3PPI.
Lu et al. (Mon,) studied this question.