I recall sitting in a lecture a few decades back discussing proteins and their ability to bind "stuff". Some of the stuff are receptors, and some are ligands, some control pathways, and some control the extracellular matrix. To go from the amino acid sequence, which we can readily get from the DNA to a complex folded protein molecule with binding properties I thought was straightforward. Not really, the more I looked.
But now a Goggle entity as noted in Nature uses a neural net to perform the task. They note:
Overall, teams predicted structures more accurately this year, compared with the last CASP, but much of the progress can be attributed to AlphaFold, says Moult. On protein targets considered to be moderately difficult, the best performances of other teams typically scored 75 on a 100-point scale of prediction accuracy, whereas AlphaFold scored around 90 on the same targets, says Moult. About half of the teams mentioned ‘deep learning’ in the abstract summarizing their approach, Moult says, suggesting that AI is making a broad impact on the field. Most of these were from academic teams, but Microsoft and the Chinese technology company Tencent also entered CASP14.
The challenge however is that proteins are not static. They can change their conformation based upon their environment, such as pH. But this is a workable first step.
Posts
