AI Solves 50-year-old Science Problem in ‘STUNNING ADVANCE’

[The IbanezerScrooge]

...Could Dramatically Change the Way We Fight Disease.

AI solves 50-year-old science problem in ‘stunning advance’ that could change the world

This is YUGE! I mean really huge. Big, big deal. I really hope the paper pans out in peer-review and this is legit. In 10-20 years we could have the cure for cancers and autoimmune diseases... maybe even the common cold!!

Really, though this is a milestone in medical history. Equivalent to when the internet was first invented.

 

[Ophiolite]

...Could Dramatically Change the Way We Fight Disease.

AI solves 50-year-old science problem in ‘stunning advance’ that could change the world

This is YUGE! I mean really huge. Big, big deal. I really hope the paper pans out in peer-review and this is legit. In 10-20 years we could have the cure for cancers and autoimmune diseases... maybe even the common cold!!

Really, though this is a milestone in medical history. Equivalent to when the internet was first invented.

I saw the announcement via a BBC article. I agree if the results are confirmed and built upon this has the potential for the dramatic consequences you mention. Nobel Prize territory? I'm not a chemist so I can only wonder.

My caution is based on two things:

• Popular media like to dramatise things
• While the software determined the shape of around two thirds of the proteins with accuracy comparable to laboratory experiments, this means around 1/3 are not determined accurately. I'm not clear how the confidence in the accuracy of any specific protein could be assessed without carrying out the laboratory testing that this is meant to largely replace. I expect that this has been considered and solutions are in hand, but until . . . .

But if it does prove out then your enthusiasm will be well justified.

 

[sesquiterpene]

I saw the announcement via a BBC article. I agree if the results are confirmed and built upon this has the potential for the dramatic consequences you mention. Nobel Prize territory? I'm not a chemist so I can only wonder.

My caution is based on two things:

• Popular media like to dramatise things
• While the software determined the shape of around two thirds of the proteins with accuracy comparable to laboratory experiments, this means around 1/3 are not determined accurately. I'm not clear how the confidence in the accuracy of any specific protein could be assessed without carrying out the laboratory testing that this is meant to largely replace. I expect that this has been considered and solutions are in hand, but until . . . .

But if it does prove out then your enthusiasm will be well justified.

Yes, 66% would be a failing grade on some tests - luckily they were grading on the curve. They also skipped some parts of the test. Anyway, here's a link to some discussion, which also gives a few extra links.
https://blogs.sciencemag.org/pipeline/archives/2020/11/30/protein-folding-2020

 

[Ophiolite]

Yes, 66% would be a failing grade on some tests - luckily they were grading on the curve. They also skipped some parts of the test. Anyway, here's a link to some discussion, which also gives a few extra links.
https://blogs.sciencemag.org/pipeline/archives/2020/11/30/protein-folding-2020

Very interesting, with several intriguing points.
I suspected something along these lines, but lacked the vocabulary and background to express it:
"such computations tend to use a variety of techniques: straight-out calculation of those energetic factors mentioned above (when necessary) along with searching for similarities to known protein sequences and structures to get a leg up. Improved methods to run such “prior art” searches reliably are a big area as well; they are nontrivial."

And in one of the links this less optimistic assessment of the technique's potential:
"Let’s consider, though, this claim . . . . “Amazon, Google, and Facebook are using AI to find protein structures. . . The tech giants are becoming increasingly active in deciphering protein structures, a crucial step in helping to find drugs” We heard a lot about how crucial this was during the last protein-folding competition as well. But is this true?
I’m going to sound like a heretic (at least to the people doing that work!) and say no, not generally. There are cases where such structures can be very helpful, but there are cases where they don’t do you that much good at all."

 

[essentialsaltes]

Does this put crystallographers, or whatever, out of work? Or rather put them in the place of 'confirming' structures devised 'in silico'?

 

[Tanj]

Does this put crystallographers, or whatever, out of work? Or rather put them in the place of 'confirming' structures devised 'in silico'?

Nope.

Ironically, it's the ever increasing body of empirical data the crystallographers put into the public domain that lets these algorithms improve. It's also important to note it showed strong improvement in the "can you recognise an apple from an orange" section of the competition, not the can you find the granny smith from the other apples in this picasso painting section, and it got to 99% in 66% of them.

It's more a strong incremental improvement than an award winning leap in understanding.

One area it will help with massively is the prediction of non synonymous mutations on single protein structures.