Here is an article about DeepMind which, over two years, has deduced the structure of millions of proteins. If accurate, it is a brilliant piece of applied physical mathematics. Is this going to be a big deal? Maybe it is, though they published 20,000 human proteins in a public database last year and it has not led to much to my limited knowledge although these things take time.
Any opinions? Involvement? Other? Is this as huge a deal as advertised?
It’s a big step. If nothing else, it’s a very difficult problem, and so the fact that a computer can do it is a significant milestone for how good computers are getting.
As for practical biological application, as I understand it, the real holy grail is to solve the problem in the other direction: Given a desired shape for the finished protein, come up with an amino acid sequence that will fold into that shape. That would enable the rapid development of all sorts of drugs, because most drugs work the way they do because they have the right shape for something-or-other.
Now, once you’ve solved the problem of finding what shape a sequence makes, you can brute-force this problem, just by making a bunch of random sequences until you find one that has the right shape. But that’s likely to be slower than you want. Though on the bright side, if everyone doing this shares their data, then even the shapes that aren’t useful for you might be useful for someone else.
My daughter, who has a master’s in biochem and is now a scientific editor thinks it is a very big deal. Knowing how proteins fold gets you a long way to understanding how they work (or don’t).
An embarrassment of riches.
Ingenious programs paired with ever more powerful computers can produce overwhelming amounts of data.
In some cases this can zero in on a solution. In other cases it can only produce a very large set of possibilities.
It depends on what the program was targeted to do.
This particular one was just meant to produce as many varieties of proteins as possible. Not to consider a particular case that required a protein and find it.
But it definitely has value as an available database of possible proteins. Maybe someone can see one of these as important to their research? Maybe another program can take this data and produce more targeted results as to the use of these proteins?
But the actual laboratory and clinical trials of a small fraction of these proteins uses will take far more time than the computers take to find them.
But, there may be another program developed…
In a related AI advancement story, Artificial Intelligence Discovers Alternative Physics. This is about Newtonian physics, doesn’t sound like time machine type stuff yet, but who knows. I think we’ll see more of this, AI systems targeted at specific problems and developing methodology from meta-analysis of random examples.
I hope so. If the results are accurate, maybe better medicines are on the way, and problems like antibiotic resistance become historical. There is potential for great advances but no shortage of purple prose that had high initial expectations.
It is an incredibly big deal. Potentially Nobel Prize winning eventually. It has already changed the way we do things in my lab.
I had a colleague years ago whose entire research program was focused around predicting protein structure from a primary amino acid sequence. I was not aware how impossible this was at the time. Usually you found a related protein that did have a known structure (from crystallography) and mapped your sequence onto it to find a “close enough” predicted struxture.
I jad a new viral protein with no good hits to the known structure database. I figured (knowing nothing about the protein structure field) “hey, you can predict DNA and RNA secondary structure from the primary sequence, why not a protein structure!”. And went looking for an online tool to do juat that.
I did not find one. So I went to my colleague and asked what tool he would recommend.
He looked at me sadly like I was a slow child 
and patiently explained the problem to me.
The discovery of this ultimately led to me deciding not to keep pushing the whole Folding@home thing on my computer when my fans started making more noise than I’d like when running at high all the time. It just is so much better by orders of magnitude, and articles I read said that F@H was already unlikely to help.
It was just considered that big a change back when it was first announced.
Caveat: I’m about 6 months behind in my literature reading and more than a decade past when I was regularly thinking about drug discovery.
As of late last year, AlphaFold’s C-alpha root mean square deviation accuracy was too high for structure-based drug design. And even if it were, that’s not the slow step is drug discovery.
IIRC the Wikipedia article on AlphaFold has more a cautious optimism than what you’ll see in press releases or the “science journalism” pieces that parrot them.
Just bumping this thread. I was, of course, correct. Although I would not have predicted a Nobel within just a couple years. Crazy.