The State of Protein Structure Prediction and Related Advancements

Explore the groundbreaking advancements in protein structure prediction through this illuminating lecture by Mohammed AlQuraishi from Columbia University. Delve into the revolutionary impact of AlphaFold2 on structural biology and its ability to accurately predict protein structures from sequences. Discover the limitations of AlphaFold2's implementation, including the lack of code and data for training new tasks, optimization issues for commercial hardware, and gaps in understanding how training data influences accuracy. Learn about OpenFold, an optimized and trainable version of AlphaFold2, and its ability to reproduce AlphaFold2's accuracy. Gain insights into the relationships between data size, diversity, and prediction accuracy, as well as how OpenFold learns to fold proteins during its training process. This talk, part of the AI≡Science series, strengthens the bond between artificial intelligence and the sciences, offering valuable knowledge for researchers and enthusiasts in the field of computational biology and protein structure prediction.