AlphaFold Catalogues 200 Million Protein Fold Predictions

[.Scott]

TL;DR

The protein predictions are based on AI models. Overall quantitative accuracy does not seem to be reported.

In their (AlphaFold) words:

AlphaFold is an AI system developed by DeepMind that predicts a protein’s 3D structure from its amino acid sequence. It regularly achieves accuracy competitive with experiment. ... The latest database release contains over 200 million entries, providing broad coverage of UniProt (the standard repository of protein sequences and annotations).

The value of this as a news story is directly proportional to the accuracy of the predictions.
The AlphaFold Database includes information about their confidence for each protein segment, but I have not found an overall quantitativ assessment of its accuracy.

 

[mfb]

Overall quantitative accuracy does not seem to be reported.

An average is probably not that meaningful as different proteins will have different importance to research.

This looks like a big breakthrough if the typical accuracy is good enough.

 

[Filip Larsen]

For what it is worth, Financial Times also picked up on that story here repeated by Ars Technica: https://arstechnica.com/science/202...acks-structure-of-almost-every-known-protein/