AI system solves 50-year-old protein folding problem in hours

Jul 27, 2020
304
47
230
Visit site
If this AI holds up at 90+ for most sequences it reads and predicts structure, it certainly is an outstanding achievement. Of course one has to know the sequence of the protein before AlphaFold can do its job. But what a job it appears to be doing. Not surprising that it doesn't do well with protein-protein structural interactions since that is not part of the folding process. But perhaps they will be able to improve on it over time.

On the experimental side of determining protein structure, what they did not tell you is that it no longer takes years and years to obtain protein structure at atomic resolution. The field of cryo-electron microscopy (1) allows for this determination much more rapidly, and is viable for various structural forms that some proteins take up which are not suitable for AlphaFold to predict. This is because the structures of some proteins might vary considerably during its activity. Such conformational changes would not be easy to establish with a sequence-only based determination. It would seem that the two approaches may work in combination, and in many instances determine a great deal about protein structure and function.

Indeed, cryo-EM is so fast, it is already providing a considerable amount of critical information about the spike protein of SARS-CoV-2 (2). The resolution of proteins by cryo-EM, which is a direct approach to its actual structure, is quite remarkable.



"Cryo-electron microscopy reaches atomic resolution"

1. https://www.nature.com/articles/d41586-020-02924-y


"Cryo-EM Structure of the 2019-nCoV Spike in the Prefusion Conformation"

2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7217118/
 
Last edited: