Two years ago, Google team made an incredible advance in structural biology, practically solving the protein folding problem (predicting protein structure from its amino acid sequence). The AI-based AlphaFold algorithm was shown to produce protein models comparable in quality to the experimental ones. It shaked up the field of structural biology, which now must contend with having a choice between spending months on experimental structure determination, or pick up a ready AlphaFold model. However, while amazingly successful, AlphaFold doesn’t address several important questions about proteins, the main one being the existence of alternative conformations. Alphafold is based on the one sequence - one structure paradigm and ignores or rejects all information not compatible with this paradigm. In this seminar I want to highlight a possible path to use some of this information and seek collaborations with deep learning experts to use the computational tools from the AlphaFold infrastructure (or to invent new ones) to address some really fascinating questions in structural biology.

Prof. Adam Godzik