In July, DeepMind announced that its Alpha Fold model had figured out how most proteins in our bodies fold.
Pushmeet Kohli told New Scientist that there is more to do. It took scientists decades to decipher the structure of only 17 percent of proteins in the human body.
But British AI firm DeepMind raised the bar to 98.5 percent in July when it announced its AlphaFold model. could quickly and reliably calculate how proteins fold.
This can lead to targeted drugs binding to specific parts of the molecule. We spoke with Pushmeet Kohli at DeepMind and saw work to map nearly every one of the more than 100 million known proteins sequenced from the entire tree of life.
Are you surprised by AlphaFold’s success, considering that protein folding previously required large supercomputers?
We present papers where machine learning and AI come into play. But many on the team weren’t sure if the problem could be solved. It was a very pleasant surprise.
They plan to release more protein structures. Why not leave the question to the scientists who can now use AlphaFold?
We’ve made the model and code public so anyone in the world can find the structure of any protein they want. We’ve seen universities and labs around the world use our code. But the reason we extend the database version is because it takes a lot of time and investment. and you don’t want different people finding the structure of the same protein over and over again, right?. It would be very useful if we actually did this once and for all for everyone.
What do you do first?
We received feedback from the community on which organisms and which types of proteins we should prioritize next. So we’re working along this roadmap, and ultimately we’re moving in the direction we’re committed to, which is to unravel the structure of the entire protein universe.
Does this mean new work or just applying AlphaFold at scale?
The team continuously improves the accuracy of the model. But we also want to extend the capabilities of AlphaFold. So we’ve been looking at individual proteins, but complexes are important because if you look at the biological mechanisms at work, it’s very rare for a single protein to interact with another small molecule in isolation. Composite Structures – We extend AlphaFold for this.
Are you going to get to a point where you’ve mapped everything and AlphaFold can retire?
Protein will change, life will change. As evolution progresses, you’ll see different types of proteins come into play. So AlphaFold will have life, not only in the complex, but in thinking about how the structure develops.
So what about Covid-19?
We discovered the structure of all SARS-CoV-2 proteins very early on. Some have been experimentally verified, but many are difficult to figure out experimentally. When scientists actually discovered these structures, it was interesting to see that ours were very consistent.
Well, for variants, there is also a factor that these small mutations cause structural changes, but AlphaFold is currently not sensitive to very small changes. So we want to make sure that future versions of AlphaFold are very sensitive to mutations.