AI to predict protein structure

In case you missed it, a highly important paper has just appeared, ‘Highly accurate protein structure prediction for the human proteome’ by Tunyasuvunakool et al.

The researchers work at an AI company, DeepMind, based in the UK. Until now, predicting the folding of a protein product from the sequences that code for it has not been possible. The DeepMind team has developed a machine-learning-based method called AlphaFold which enables this prediction with high accuracy. To date, hundreds of thousands of shapes have been predicted from sequences.

A brief overview is here:

https://deepmind.com/research/publications/2021/enabling-high-accuracy-protein-structure-prediction-at-the-proteome-scale

The method is described here:

https://www.nature.com/articles/s41586-021-03819-2

And its large-scale application to the human genome is here:

https://www.nature.com/articles/s41586-021-03828-1

Note that the breakthrough came from the computing world, but the application is bioscience. Moreover, the AlphaFold application itself was not ‘taught’ by experts but learned from examples. It represents a major realignment in research direction and priorities.

The ability to predict what shape a particular gene sequence will give to a protein ‘in silico’ is new and huge, and will have substantial impact across nearly every life sciences field.