Ask HN: What does protein folding have to do with OpenAI and ChatGPT?

2 points·theturtletalks··4 comments

I saw a link here about the most cited AI academic articles and almost all were about protein folding. Why was solving protein folding so vital in the AI world?

View on HN

ML has not solved protein folding. What it did do was dramatically advance the state of the art of protein structure prediction, a much easier problem, and not really solved.

In principle, being able to predict protein structures with high accuracy could greatly enhance the rate at which we can understand human diseases and come up with treatments and cures. In practice.... that's not really true.

BUt it's still nice to be able to point to a specific ML model and say "this 30+ year problem was substantially addressed"; it shows people banged their heads on something for 30 years, got stuck, and finally ML found a nice way out

In a literal sense in answer to your question: nothing.

Language models have been used by Meta and others to enhance the protein fold work of Alphafold2. Conjoined with diffusion networks, they're able to generate de novo proteins[0].

OpenAI has nothing going [publicly] in the burgeoning generative protein space.

[0] https://generatebiomedicines.com/chroma

It has a large number of potential scientific/medical applications.

Yes, but the more I read about it, solving protein folding was a major breakthrough for all AI applications.

Reply on news.ycombinator.com