Fine, I'll Take The Bait

Ok.

I wasn't going to, but when both Charlie Stross and Moe Lane brought up this article from Google/DeepMind/Alpha Fold about protein folding methods... well I couldn't quite stay away.

Putting on my cynical hat for a second, I'll first point out that at least one of the reasons the Alpha Fold story is getting pushed out to the web is because of this article by Emiliano Brini, Carlos Simmerling, and Ken Dill in this week's edition of Science.

If you don't have a Science/AAAS membership, the way they do the abstracts for their headline stories these days does a good job of summarizing the whole thing; if you can read the paper, you'll note that Alpha Fold gets a brief mention somewhere around paragraph 36 (of 37).

I don't point that out to dump on Alpha Fold. They're doing good work at Deep Mind.

However, I'll point out that Ken Dill and company paint a much broader picture here, one that sets Alpha Fold into an ensemble of techniques, and properly so.

The key element here is that which Dill and company stress: Dynamics. (If you really want an earful here, wait until Courtney Milan recovers from the stress of the moment and then ask her. She did her masters across the bay from where Ken Dill used to be based at UCSF; I can promise you that Courtney and her old mentor David Chandler had insights into this area, it's a big part of what David worked on in his latter years.)

Levinthal's Paradox still very much governs here. And we've learned so much; there are so very many proteins which turn out not to have a structure that is accessible via crystallization. This is why CryoEM is such a big deal. CryoEM means we can capture a structural picture of proteins that are not otherwise accessible to XRay crystollagraphy because of their lability.

Personally? I think this article at C&E News on cheap CryoEM is just as interesting, if not more so, than the Alpha Fold story. A million dollar CryoEM means on-site structural work to approximately 2 angstrom resolution at pretty much *any* good-sized lab. Combine that with a good computational cluster of the same order of expense, and we're talking a level of biomolecular investigation that will absolutely combine to solve problems us old farts only dreamt of. And it will only get better and cheaper, unlike beam time.

Summary? Oh boy howdy have we come a long way in understanding the computational details of proteins. But oh so very far have we to yet go. If you're excited about Alpha Fold, clap for them.

And then think about what it means that they're only a small piece of a very large, very complex puzzle. In 6N dimensions, where N is Avogadro's number. There are marvels here awaiting us.