Protein folding solved?

Get your science fix here: research, quackery, activism and all the rest
User avatar
bolo
Snowbonk
Posts: 565
Joined: Mon Nov 11, 2019 1:17 pm
Location: Washington DC

Re: Protein folding solved?

Post by bolo » Thu Dec 03, 2020 3:56 am

Bird on a Fire wrote:
Thu Dec 03, 2020 12:01 am
Yeah I feel like stuff like physics and space get way more huge public projects funded - all those space missions and particle accelerators and supercomputers and things. Biology seems to be expected to survive to a much greater extent as a by-product of commercial research. (I haven't checked the numbers, but I can't think of many similar big public biology projects - even the coronavirus vaccine has been largely funnelled through private corporations, and that was a global emergency)
In the United States, the National Institutes of Health spends far more on basic and applied research than any other federal agency. About triple the second-place agency (NASA) on basic, and more than double the second-place agency (DOE) on applied.

What NIH doesn't spend on is development, on which DOD and NASA and DOE spend billions. That's why U.S. COVID-19 vaccine development was done largely by private companies, though still with lots and lots of federal money (most of it in the form of advance purchases, rather than research grants).

I'm not a biologist, and I don't follow NIH all that closely, so I can't claim to understand why it has chosen not to fund more big projects, but given that it's relatively awash in money, my starting guess would mostly be that it doesn't think that's the most effective way to spend it. Possibly also that it doesn't have much experience with running such projects -- unlike DOE, for example, which has lots of big facilities, and does biology with some of them (e.g. supercomputers and light sources as mentioned above) even though someone who just saw the words "Department of Energy" might not expect that.

User avatar
bjn
Catbabel
Posts: 951
Joined: Wed Sep 25, 2019 4:58 pm

Re: Protein folding solved?

Post by bjn » Thu Dec 03, 2020 8:43 am

bolo wrote:
Wed Dec 02, 2020 10:35 pm
If Google weren't a behemoth corporation, it wouldn't have behemoth-level money to throw at this.

If this had to be free and open-source, Google wouldn't expect to make a profit on it and wouldn't want to throw money at it.

You can call the researchers massively naive about the economic implications. You might find that they'd call you massively naive about the alternative.
??? Your very next post in this thread makes pretty much the same points I made, but with more detail. Personally, I think critical technologies like this shouldn't be locked away in private companies. You seemed to be thinking I was making a value judgement.

User avatar
Boustrophedon
After Pie
Posts: 1642
Joined: Mon Nov 11, 2019 3:58 pm
Location: Lincolnshire Wolds

Re: Protein folding solved?

Post by Boustrophedon » Thu Dec 03, 2020 10:44 am

dyqik wrote:
Wed Dec 02, 2020 12:12 am
A fairly major issue with this is that it's a closed project that other researchers can't use (and is built on open info/tools).

As such, as far as doing science is concerned, it doesn't exist.
Worse than that it is not predictive as the AI, based on neural nets does not inform as to the why proteins fold as they do so it adds nothing to the theory. A useful empirical tool perhaps, but that's all.
Remember it's only a coup if it's from the coup d'état region of France, otherwise it just sparkling white terrorism.

User avatar
dyqik
Stummy Beige
Posts: 3247
Joined: Wed Sep 25, 2019 4:19 pm
Location: Masshole
Contact:

Re: Protein folding solved?

Post by dyqik » Thu Dec 03, 2020 1:25 pm

bolo wrote:
Thu Dec 03, 2020 3:29 am
dyqik wrote:
Wed Dec 02, 2020 11:32 pm
A Government funded facility for protein folding calculations using this technology would do protein folding calculations. In the same way that a government funded facility for radio astronomy does radio astronomy rather than satellite tracking and imaging.
A government funded facility for protein folding calculations would look a lot like a government funded facility for any other high-performance scientific computing task, e.g. the National Energy Research Scientific Computing Center at Lawrence Berkeley National Lab, or the Leadership Computing Facilities at Oak Ridge National Lab and Argonne National Lab. All of these have funded at least some work on protein folding. They all regularly fund biology -- COVID-related research, for example, has been a big push for them this year. But they also fund a vast array of other topics, and that broad base is part of why it was attractive for the government to fund them.
It'd look like them, but it'd be funded to do protein folding more extensively. It'd mean dedicating money to operations of such a center (with Google being paid for their tech), and designing it to suit the Google tech rather than general supercomputing problems, in exchange for a guaranteed slice of the time for that work.

I do know about the computing facilities. We use the Smithsonian's one for work I do, and the Argonne one for CMB work.

User avatar
jdc
Hilda Ogden
Posts: 678
Joined: Wed Sep 25, 2019 4:31 pm
Location: Your Mum
Contact:

Re: Protein folding solved?

Post by jdc » Thu Dec 03, 2020 8:25 pm

bjn wrote:
Thu Dec 03, 2020 8:43 am
bolo wrote:
Wed Dec 02, 2020 10:35 pm
If Google weren't a behemoth corporation, it wouldn't have behemoth-level money to throw at this.

If this had to be free and open-source, Google wouldn't expect to make a profit on it and wouldn't want to throw money at it.

You can call the researchers massively naive about the economic implications. You might find that they'd call you massively naive about the alternative.
??? Your very next post in this thread makes pretty much the same points I made, but with more detail. Personally, I think critical technologies like this shouldn't be locked away in private companies. You seemed to be thinking I was making a value judgement.
Wasn't Bolo's comment aimed at Boaf - it was him who called the researchers "massively naive".

User avatar
bjn
Catbabel
Posts: 951
Joined: Wed Sep 25, 2019 4:58 pm

Re: Protein folding solved?

Post by bjn » Thu Dec 03, 2020 8:34 pm

Reading fail. Looked like an strange reply to mine.

Beaker
Buzzberry
Posts: 45
Joined: Tue Nov 12, 2019 5:28 pm

Re: Protein folding solved?

Post by Beaker » Fri Dec 04, 2020 12:13 am

Boustrophedon wrote:
Thu Dec 03, 2020 10:44 am
dyqik wrote:
Wed Dec 02, 2020 12:12 am
A fairly major issue with this is that it's a closed project that other researchers can't use (and is built on open info/tools).

As such, as far as doing science is concerned, it doesn't exist.
Worse than that it is not predictive as the AI, based on neural nets does not inform as to the why proteins fold as they do so it adds nothing to the theory. A useful empirical tool perhaps, but that's all.
It is indeed a useful tool - as in could be potentially game changing for developing medicines if you can go from a protein sequence or even a DNA sequence to a structure in-silico without trying for a year to grow a crystal.

You could electronically evolve a protein to be a bio catalyst for a specific reaction, build up libraries of structures and put them in the public domain for orders of magnitude more molecules than currently available...

No, probably doesn’t advance the ab-initio calculations very far. Am I bovvered?

User avatar
Grumble
Dorkwood
Posts: 1534
Joined: Mon Nov 11, 2019 1:03 pm

Re: Protein folding solved?

Post by Grumble » Fri Dec 04, 2020 8:38 am

If I was interested in a particular protein, let’s say I asked Deep Mind and they sent me a structure, how could I confirm that it’s right? Does having a candidate structure make it easier to do the crystallography or whatever? They claim about 90% accuracy so clearly there is a need to check.

If it makes it easier to sort out the next step it’s great, but if it needs checking and doesn’t help the checking process then we’re no further forward.
I know this is vitriol, no solution, spleen venting, but I feel better having screamed, don’t you?

Beaker
Buzzberry
Posts: 45
Joined: Tue Nov 12, 2019 5:28 pm

Re: Protein folding solved?

Post by Beaker » Fri Dec 04, 2020 9:07 am

Each technique has its error bars for the position of every atom. In the competition that’s making the news, they compared AI results to recent experimentally determined structures, as you suggest. The level of accuracy compared well to the experimentally determined structures. Any new analytical technique should certainly be validated against other methods, but it does look like it has merit on its own.

And yes, solving an x-ray crystal structure would be easier if you have a good idea of what it looks like to start with.

Edit: and getting a crystal of a protein to X-ray is by no means guaranteed

User avatar
Bird on a Fire
Light of Blast
Posts: 4849
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: Protein folding solved?

Post by Bird on a Fire » Fri Dec 04, 2020 12:42 pm

Boustrophedon wrote:
Thu Dec 03, 2020 10:44 am
dyqik wrote:
Wed Dec 02, 2020 12:12 am
A fairly major issue with this is that it's a closed project that other researchers can't use (and is built on open info/tools).

As such, as far as doing science is concerned, it doesn't exist.
Worse than that it is not predictive as the AI, based on neural nets does not inform as to the why proteins fold as they do so it adds nothing to the theory. A useful empirical tool perhaps, but that's all.
This is one of the frustrations of AI and machine-learning approaches in general - they often do really well at predicting stuff, but trying to understand how they make their predictions is almost like reading entrails: it'll be a massively opaque tangle of multivariate maths. The AI more or less understands why proteins fold as they do, but can't communicate that information to the humans.

OTOH, this might make it cheaper and easy to get much bigger (unverified) samples, perhaps even including proteins that don't exist in reality, to try to explore how proteins work through in silico experiments, so it can still be used predictively/to test theories.
THINK BIG AND UPEND THE SYSTEM

#ShowYourStripes

monkey
Fuzzable
Posts: 355
Joined: Wed Nov 13, 2019 5:10 pm

Re: Protein folding solved?

Post by monkey » Fri Dec 04, 2020 4:27 pm

On checking:

If you've got a model of a protein folding, might there be easier tests to confirm that the model is correct than X-ray diffraction? That would seem useful to me. My brother in law might be annoyed though, he grows the crystals for a living.

User avatar
bolo
Snowbonk
Posts: 565
Joined: Mon Nov 11, 2019 1:17 pm
Location: Washington DC

Re: Protein folding solved?

Post by bolo » Fri Dec 04, 2020 6:55 pm

In case anyone's interested, legislation that will almost certainly pass Congress in the next few days (the annual defense authorization act) includes direction to the Department of Energy regarding government-funded computing facilities for AI:
(d) FACILITY USE AND UPGRADES.—In carrying out the program under subsection (a), the Secretary shall—
(1) make available high-performance computing infrastructure at national laboratories;
(2) make any upgrades necessary to enhance the use of existing computing facilities for artificial intelligence systems, including upgrades to hardware;
(3) establish new computing capabilities necessary to manage data and conduct high performance computing that enables the use of artificial intelligence systems; and
(4) maintain and improve, as needed, networking infrastructure, data input and output mechanisms, and data analysis, storage, and service capabilities.

User avatar
Bird on a Fire
Light of Blast
Posts: 4849
Joined: Fri Oct 11, 2019 5:05 pm
Location: nadir of brie

Re: Protein folding solved?

Post by Bird on a Fire » Sat Dec 05, 2020 8:57 pm

This came up on twitter - it does seem that AI research at Big Tech companies is directly competing with universities.
Screenshot at 2020-12-05 20-53-58.png
Screenshot at 2020-12-05 20-53-58.png (218.42 KiB) Viewed 152 times
chart from https://docs.google.com/presentation/d/ ... 896d_0_102
tweet: https://twitter.com/ubiquity75/status/1 ... 6508469251

A lot of people seem to feel uneasy about these changes.
THINK BIG AND UPEND THE SYSTEM

#ShowYourStripes

User avatar
science_fox
Clardic Fug
Posts: 213
Joined: Mon Nov 11, 2019 1:34 pm
Location: Manchester

Re: Protein folding solved?

Post by science_fox » Sat Dec 05, 2020 9:25 pm

monkey wrote:
Fri Dec 04, 2020 4:27 pm
On checking:

If you've got a model of a protein folding, might there be easier tests to confirm that the model is correct than X-ray diffraction? That would seem useful to me. My brother in law might be annoyed though, he grows the crystals for a living.
Along with all the arguments about how representative a contrived static snapshot actually is compared to what happens in the biology. Few other techniques can pinpoint atom positions though, but how they move is just as important.

I wonder how well it models Intrinsically Disordered Proteins that don't have a structure as such, but occupy a variety of energy minima?

User avatar
bolo
Snowbonk
Posts: 565
Joined: Mon Nov 11, 2019 1:17 pm
Location: Washington DC

Re: Protein folding solved?

Post by bolo » Fri Dec 11, 2020 9:07 pm

bolo wrote:
Fri Dec 04, 2020 6:55 pm
In case anyone's interested, legislation that will almost certainly pass Congress in the next few days (the annual defense authorization act) includes direction to the Department of Energy regarding government-funded computing facilities for AI:
(d) FACILITY USE AND UPGRADES.—In carrying out the program under subsection (a), the Secretary shall—
(1) make available high-performance computing infrastructure at national laboratories;
(2) make any upgrades necessary to enhance the use of existing computing facilities for artificial intelligence systems, including upgrades to hardware;
(3) establish new computing capabilities necessary to manage data and conduct high performance computing that enables the use of artificial intelligence systems; and
(4) maintain and improve, as needed, networking infrastructure, data input and output mechanisms, and data analysis, storage, and service capabilities.
The House and Senate have now passed this bill. Trump has threatened to veto it (for reasons completely unrelated to this provision) but it's unclear whether he really will, and if he does, there's a good chance Congress will override him.

AMS
Snowbonk
Posts: 466
Joined: Mon Nov 11, 2019 11:14 pm

Re: Protein folding solved?

Post by AMS » Sat Dec 12, 2020 6:13 pm

bolo wrote:
Thu Dec 03, 2020 3:56 am
Bird on a Fire wrote:
Thu Dec 03, 2020 12:01 am
Yeah I feel like stuff like physics and space get way more huge public projects funded - all those space missions and particle accelerators and supercomputers and things. Biology seems to be expected to survive to a much greater extent as a by-product of commercial research. (I haven't checked the numbers, but I can't think of many similar big public biology projects - even the coronavirus vaccine has been largely funnelled through private corporations, and that was a global emergency)
In the United States, the National Institutes of Health spends far more on basic and applied research than any other federal agency. About triple the second-place agency (NASA) on basic, and more than double the second-place agency (DOE) on applied.

What NIH doesn't spend on is development, on which DOD and NASA and DOE spend billions. That's why U.S. COVID-19 vaccine development was done largely by private companies, though still with lots and lots of federal money (most of it in the form of advance purchases, rather than research grants).

I'm not a biologist, and I don't follow NIH all that closely, so I can't claim to understand why it has chosen not to fund more big projects, but given that it's relatively awash in money, my starting guess would mostly be that it doesn't think that's the most effective way to spend it. Possibly also that it doesn't have much experience with running such projects -- unlike DOE, for example, which has lots of big facilities, and does biology with some of them (e.g. supercomputers and light sources as mentioned above) even though someone who just saw the words "Department of Energy" might not expect that.
There's quite a lot of "big charity" funding in Biology too, particularly for anything in the human health sector. Wellcome, Gates Foundation for example (both of which were founded by wealthy philanthropists), as well as lots of disease-specific charity funding from the likes of CRUK. Wellcome were a big part to the human genome project (the Sanger Centre is a Wellcome institution) while Gates have put a lot into neglected diseases.

AMS
Snowbonk
Posts: 466
Joined: Mon Nov 11, 2019 11:14 pm

Re: Protein folding solved?

Post by AMS » Sat Dec 12, 2020 6:24 pm

Also, NIH run the NCBI (National Centre for Biotechnology Information), which together with its European and Japanese equivalents, run all the big public databases of genes/proteins.

Post Reply