Computational Advances in Structural Biology

Deep learning is revolutionizing the prediction of protein tertiary and quaternary structures. In this talk, I will first describe how we significantly improved the performance of the state-of-the-art deep learning-based protein structure prediction method – AlphaFold2 in our MULTICOM protein structure prediction system, which ranked among the top tertiary and quaternary structure prediction servers in the 15th Critical Assessment of Techniques for Protein Structure Prediction (CASP15) in 2022. I will then present our latest work of applying 3D-equivariant graph transformers with the self-attention to refine protein structural models. Our experiments demonstrate that 3D-equivariant graph transformers that are robust against the rotation and translation of 3D objects can improve the quality of protein structures more effectively than the existing methods.