Post 26: 600 Million mew protein tructures! 🤯

One can wake up any day and find out that there are now 600 million new protein structures predicted using artificial intelligence. Of these, ~225 million are predicted with high confidence, and about 12% have no record with any other known structure. All thanks to the use of ~2000 GPUs running for 2 weeks. Quite casual.

Refs:

1. ESMAtlas
2. A nature comment
3. (the paper) Evolutionary-scale prediction of atomic level protein structure with a language model
4. More about protein language models
5. More about the model ESM-1b
6. The 200M models predicted with AlphaFold2