"Scores above 90 on the zero to 100 scale are considered on par with experimental methods, Moult says." Who is it that does the considering for us? Great that it's good enough to make molecular replacement work (VERY great!!!!) - but "on par" is a big word. Sent from tiny silly touch screenhttp://www.9folders.com/ ________________________________ From: Jim Fairman Sent: Monday, 30 November 2020 19:58 To: lbetts0508 Cc: PHENIX user mailing list Subject: Re: [phenixbb] alpha-Fold 2? For the most challenging proteins, AlphaFold scored a median of 87, 25 points above the next best predictions. It even excelled at solving structures of proteins that sit wedged in cell membranes, which are central to many human diseases but notoriously difficult to solve with x-ray crystallography. Venki Ramakrishnan, a structural biologist at the Medical Research Council Laboratory of Molecular Biology, calls the result “a stunning advance on the protein folding problem.” Source: https://www.sciencemag.org/news/2020/11/game-has-changed-ai-triumphs-solving... -------------------------------------------------------------------------- Jim Fairman C: 1-240-479-6575 On Mon, Nov 30, 2020 at 10:25 AM lbetts0508 mailto:[email protected]> wrote: all - I just read the blurb in Nature Briefing about the DeepMind AI having made a big advance in the CASP protein fold prediction. Does it sound really transformational, does it work for membrane proteins - all the usual questions come to mind. Do we know enough yet about it? Signed an old protein crystallographer, L. Betts _______________________________________________ phenixbb mailing list [email protected]mailto:[email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]mailto:[email protected]

I think the term "experimental methods" is doing a lot of work here... ________________________________ From: [email protected] on behalf of Frank Von Delft Sent: 30 November 2020 20:22 To: [email protected] ; lbetts0508 Cc: PHENIX user mailing list Subject: Re: [phenixbb] alpha-Fold 2? "Scores above 90 on the zero to 100 scale are considered on par with experimental methods, Moult says." Who is it that does the considering for us? Great that it's good enough to make molecular replacement work (VERY great!!!!) - but "on par" is a big word. Sent from tiny silly touch screenhttp://www.9folders.com/ ________________________________ From: Jim Fairman Sent: Monday, 30 November 2020 19:58 To: lbetts0508 Cc: PHENIX user mailing list Subject: Re: [phenixbb] alpha-Fold 2? For the most challenging proteins, AlphaFold scored a median of 87, 25 points above the next best predictions. It even excelled at solving structures of proteins that sit wedged in cell membranes, which are central to many human diseases but notoriously difficult to solve with x-ray crystallography. Venki Ramakrishnan, a structural biologist at the Medical Research Council Laboratory of Molecular Biology, calls the result “a stunning advance on the protein folding problem.” Source: https://www.sciencemag.org/news/2020/11/game-has-changed-ai-triumphs-solving... -------------------------------------------------------------------------- Jim Fairman C: 1-240-479-6575 On Mon, Nov 30, 2020 at 10:25 AM lbetts0508 mailto:[email protected]> wrote: all - I just read the blurb in Nature Briefing about the DeepMind AI having made a big advance in the CASP protein fold prediction. Does it sound really transformational, does it work for membrane proteins - all the usual questions come to mind. Do we know enough yet about it? Signed an old protein crystallographer, L. Betts _______________________________________________ phenixbb mailing list [email protected]mailto:[email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]mailto:[email protected]

Depending on the target (i.e., free modeling, template based, etc.) the score follows a slightly different formula IIRC, but the major component is heavyatom GDT -- the assessors average the GDT evaluated with cutoffs at 1, 2, 4, and 8Å. I think it's 0-100 instead of 0-1 for reading ease more than anything else. On Mon, Nov 30, 2020 at 6:54 PM Aaron Oakley wrote:

Does “score" correlate with some estimate of the RMSD with respect to true structure?

–å

On 1 Dec 2020, at 7:22 am, Frank Von Delft wrote:

"Scores above 90 on the zero to 100 scale are considered on par with experimental methods, Moult says."

Who is it that does the considering for us? Great that it's good enough to make molecular replacement work (VERY great!!!!) - but "on par" is a big word.

Sent from tiny silly touch screen http://www.9folders.com/ ------------------------------ *From:* Jim Fairman *Sent:* Monday, 30 November 2020 19:58 *To:* lbetts0508 *Cc:* PHENIX user mailing list *Subject:* Re: [phenixbb] alpha-Fold 2?

For the most challenging proteins, AlphaFold scored a median of 87, 25 points above the next best predictions.* It even excelled at solving structures of proteins that sit wedged in cell membranes, which are central to many human diseases but notoriously difficult to solve with x-ray crystallography.* Venki Ramakrishnan, a structural biologist at the Medical Research Council Laboratory of Molecular Biology, calls the result “a stunning advance on the protein folding problem.”

Source: https://www.sciencemag.org/news/2020/11/game-has-changed-ai-triumphs-solving... -------------------------------------------------------------------------- Jim Fairman C: 1-240-479-6575

On Mon, Nov 30, 2020 at 10:25 AM lbetts0508 wrote:

...

all - I just read the blurb in Nature Briefing about the DeepMind AI having made a big advance in the CASP protein fold prediction.

Does it sound really transformational, does it work for membrane proteins - all the usual questions come to mind.

Do we know enough yet about it?

Signed an old protein crystallographer, L. Betts _______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]

Thank you Randy. Its good to hear that perspective. Since my initial post, this story was the BBC headline story in their online news app. At least something from the world of science is there instead of crazy politicians, whether or not there is some amount of hyperbole. Laurie Betts On Tue, Dec 1, 2020, 3:43 AM Randy John Read wrote:

Hi,

I’ve been following most of these talks. John Moult based the phrase “on par with experimental results” on the observation that homology modelling with the closest homologues hits GDT scores around 90-95. Extrapolating to the limit of an identical model doesn’t change that much, and it fits with what we know since the 1980s from the work of Chothia & Lesk that even accurate repeated structure determinations of the identical protein tend to disagree around the 0.4-0.8 Å rms level. As we’ve learned since (including work by Rob Oeffner and Kaushik Hatti in the Phaser team), bigger proteins tend to have somewhat bigger deviations, even at the identical sequence level. John might have been exaggerating slightly — “nearly on par” is probably a better representation of the achievement, but it’s still really impressive. To put it in context, even though the GDT measures aren’t the same as rms, Andriy Kryshtafovych made an attempt to translate, and said that GDT-HA values around 90-95 are roughly in the range of 1-1.4 Å rms.

By the way, in CASP all of these comparisons *are* based on the known experimental structure, which is not publicly available to the predictors but which is available to the assessors. In fact, in a few cases the people who contributed targets had been too optimistic about when they would have a structure and were only able to complete the structure determinations after they were given predicted models to use in MR, generally the AlphaFold models! One of these was, indeed, a membrane protein.

Randy Read

...

On 1 Dec 2020, at 00:16, Andy Watkins wrote:

Depending on the target (i.e., free modeling, template based, etc.) the score follows a slightly different formula IIRC, but the major component is heavyatom GDT -- the assessors average the GDT evaluated with cutoffs at 1, 2, 4, and 8Å. I think it's 0-100 instead of 0-1 for reading ease more than anything else.

On Mon, Nov 30, 2020 at 6:54 PM Aaron Oakley wrote: Does “score" correlate with some estimate of the RMSD with respect to true structure?

–å

...

On 1 Dec 2020, at 7:22 am, Frank Von Delft wrote:

"Scores above 90 on the zero to 100 scale are considered on par with experimental methods, Moult says."

Who is it that does the considering for us? Great that it's good enough to make molecular replacement work (VERY great!!!!) - but "on par" is a big word.

Sent from tiny silly touch screen From: Jim Fairman Sent: Monday, 30 November 2020 19:58 To: lbetts0508 Cc: PHENIX user mailing list Subject: Re: [phenixbb] alpha-Fold 2?

For the most challenging proteins, AlphaFold scored a median of 87, 25 points above the next best predictions. It even excelled at solving structures of proteins that sit wedged in cell membranes, which are central to many human diseases but notoriously difficult to solve with x-ray crystallography. Venki Ramakrishnan, a structural biologist at the Medical Research Council Laboratory of Molecular Biology, calls the result “a stunning advance on the protein folding problem.”

Source: https://www.sciencemag.org/news/2020/11/game-has-changed-ai-triumphs-solving...

---

...

...

Jim Fairman C: 1-240-479-6575

On Mon, Nov 30, 2020 at 10:25 AM lbetts0508 wrote: all - I just read the blurb in Nature Briefing about the DeepMind AI having made a big advance in the CASP protein fold prediction.

Does it sound really transformational, does it work for membrane proteins - all the usual questions come to mind.

Do we know enough yet about it?

Signed an old protein crystallographer, L. Betts _______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected] _______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected] _______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]

----- Randy J. Read Department of Haematology, University of Cambridge Cambridge Institute for Medical Research Tel: +44 1223 336500 The Keith Peters Building Fax: +44 1223 336827 Hills Road E-mail: [email protected] Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]