Prof. Paolo De Los Rios

Protein folding occurs spontaneously, as shown to us by Anfinsen, because the folding instructions are encoded in the amino-acid sequence.

Yet sequences from the same protein family fold to the same structure despite their sometimes-large differences, hinting at the possibility of learning. The folding rules by comparing many different instances (think of the Rosetta stone, which allowed understanding hieroglyphs).

Confident in the correctness of these idea, bioinformaticians first, and then physicists, developed a series of methods to extract these rules, culminating recently, upon intervention of computer scientists, in the remarkable success of AlphaFold (AF).

After a brief review of the history of the methods, I will show some pre-AF successful applications. I will then show how there is a natural convergence of deep (but not necessarily so deep) learning methods and ideas from statistical mechanics and how we can exploit these still simple approaches

as a discovery tool.