Post 8: A problem of quantities, structures vs. sequences 🚨

I was rereading a preprint from March of last year where they mentioned that, given the scarcity of protein structures, it was wise to learn from sequences. However, that was in the era before AlphaFold2, and in theory, in a few months, the DeepMind team will publish their results with protein models with atomic precision for approximately 100 million proteins. And this graph will look very different. Sometimes science and the way it’s done change, and very drastically. And it’s exciting to witness it live.

But what will we say in a few years? “With the advent of AlphaFold2”? Of artificial intelligence? Transformers?

Refs:

1. ProGen: Using AI to Generate Proteins