Question AI Protein Folding... F@H obsolete?

[VirtualLarry]

https://deepmind.com/blog/alphafold/

 

[ericlp]

My take would be, they are using AI to assemble the proteins and look at it in different ways... With a lot more accuracy than a human scientist could ever do. But, they will still need huge processing power to crunch the data. I don't think it will become obsolete, I think AI can more efficiently (with time) direct the path where higher chance of discovery.

AI isn't just ... directing folding but it is used for seti as well, in fact........ Some believe that we must create AI, so we can communicate with ET life, if we ever do wind up finding a signal. Also... All bets are off, when quantum computing comes on line. That's some scary arithmetic that is crazy fast. Tho, I believe that once we have a super fast quantum processor, AI will truly take off.

Will be pretty interesting in the next 5~10 years.

 

[StefanR5R]

Thanks for the interesting link. As far as I understand, this doesn't outright obsolete molecular dynamics analyses, but (1.) reduces the range of models to investigate, (2.) also speeds up the iteration within the analysis. (With my very limited understanding of the matter, and the article being very brief and sparse on detail, I may have misunderstood or missed something though.)

PS,
the idea of applying neural networks at the design & analysis of structures (in the most general sense) is old. Older than the fashion trend of putting "AI" instead of "neural network" into headlines, I suspect.

 

[StefanR5R]

PS,
while the general idea may be old, the feasibility of applying this in a field like protein folding may be recent, and the actual implementation certainly novel.

Incidentally, the recent news item from the OpenZika project mentions the use of machine learning methods as well, as a means to select compounds for further testing.

 

[kowalabearhugs]

GPUGrid's Quantum Chemistry WUs are also being used for machine learning...
https://www.gpugrid.net/forum_thread.php?id=4785&nowrap=true#51040

 

[TennesseeTony]

Heya Matt! Long time no see, friend!

Thanks for the link, Larry!

 

[kowalabearhugs]

Happy Holidays, Tony! Toni! Toné! ... and to all crunching DC @ AT!

 

[biodoc]

https://deepmind.com/blog/alphafold/

While alphafold did quite well in the CASP13 competition, prediction of protein structures using computational methods still has a long way to go IMO. Any serious protein structure studies such as protein engineering (Rosetta) and molecular dynamics (F@H and GPUGRID) require highly accurate protein structures derived from experimental methods such as x-ray crystallography.

Hopefully the alphafold scientists will publish their work in the near future. Since the work was done in a company, there is a possibility they won't publish any time soon or until they determine if they have intellectual property they can protect by filing for patents.