Why Fold? AI solves it.

[momus_98]

A month or two ago, in a general conversation with my IT coworkers about distributed computing, I mentioned that I have been folding since 2002 and my reasons for supporting the project. One of them chimed in with [paraphrasing] "Why? Those projects have moved to AI, so there's no need to do that."

Some context: many of my coworkers could reasonably be considered millennials, or even Gen Z, so their perspective differs from mine (Gen X).

Is this true? In 2024 vs. 2002 (when I joined), is the landscape markedly different? Should I bother? Lots of things have advanced in the 20+ years of the F@H project.

Should I keep going?

[BobWilliams757]

From my understanding, no AI does NOT solve it. AI can and has been used to help locate potential areas of research, specific area and timestamps that might allow drug binding., and other such things. Only more advanced simulations can confirm or deny the specifics of what happens when proteins fold.

From some of the stuff I've read AI can help lead us to potential areas, and save on research time, by things such as identification of potential cryptic pockets and such. But it takes more advanced folding simulations to determine if in fact that cryptic pocket exists to any level that can be exploited enough to allow drug binding.


Broken down, any AI or machine learning is just that. It cannot learn what is not known, but we are now realizing it might help point us in the right direction. Using AI for drug testing might speed things up. Using AI followed by folding or other simulations would speed it up even more. If money was no object they could simply try more of the 100's of millions of compounds real world testing. But even with the flaws that exist, the various computational models speed up the process.

[Joe_H]

Not all projects have moved to AI, only some. So their statement was incorrect from the start, probably based on a misunderstanding of what AI can do and what distributed computing projects can use it for.

In the case of folding, AI may be useful in finding folded states. But what it won't show are the paths followed to reach those folded states. What the simulations done through folding show are multiple paths to reach various stable folded states, and can show the interactions between those states and outside proteins and other chemical compounds such as drugs. Additional statistics get generated about which paths are more likely to be followed.