Deepmind and what it means for the future of folding
primesuspect Beepin n' BoopinDetroit, MI Icrontian
What do you suppose this means for the future of the project? Are we using a stone-age approach to building a space station? Are we carrying away the beach one grain of sand at a time?
When I first read that news my big take away was they figured out how to do correct computer modeling of the protein. If they can figure out how to translate that into a more standardized model everyone can run then no. Even doing the work in the lab is super hard. f@h resimulating proteins already done and giving confirmation to existing work would be huge since experimental sciences are very hard to reproduce. Real impact on our lives I do not believe f@h has an effective ROI but I still do it for now.
Deepmind is a powerful (most powerful?) AI but it is still only entity. Science is obsessed with reproducing results until they can get to 5 sigma (99.999%). In the near term, F@H can help reach that level by targeting the proteins that Deepmind had solved. In the long term, with Deepmind cracking the "code", I hope it will lead to huge leap in more efficient and powerful algorithms on protein folding. Then Stanford and F@H can piggyback on that knowledge and unleash a new wave in Folding that sees the project knock out entire sequences of proteins in a few dozen months instead of dozens of years. I mean, almost 2 million GPU and CPU cores currently working with F@H, we are still a powerhouse.
My understanding of the FAH research is that they're specifically interested in understanding circumstances that lead to misfolded proteins. The DeepMind work seems like a complementary development.