Hi James, One really interesting (and to me surprising) aspect of how well AlphaFold2 does is that it does really well without actually understanding chemistry and physics. (John Jumper from DeepMind talked about choices of deep learning model types and how they affect the “inductive bias” that allows things like chemistry to be learned indirectly, but it’s not programmed in.) The best example is the fantastic models they made of monomers from trimeric proteins. The monomers can be assembled into trimers that look very much like the real thing, but they really modelled just monomers — somehow the machine learning algorithm implicitly knows about the trimers from the existence of distant homologues in the PDB. As Joana said, AlphaFold2 did very well even on targets with no identifiable homologues, but I suspect that targets like these trimers will still require the presence of homologues. Anyway, the modelled monomer makes no sense as a monomer, and any sensible force field would much prefer something else that buries more surface area!
Following up on some other comments, AlphaFold2 is a pretty complete reinvention compared to the original AlphaFold from 2 years ago. AlphaFold followed a two-step process, where probability distributions for distances were learned in the first step (similar to the co-evolution constraints inferred by algorithms like the ones from Marks & Sander), and then those distance distributions were used in a minimisation step to fold the protein. If I recall, the first step used a convolutional deep neural network. In AlphaFold2, it’s all done in one end-to-end process going from sequence (and multiple sequence alignments) to xyz coordinates. The model type has changed to something called an attention module, which John Jumper said acts to implicitly and iteratively learn a graph representing atoms and their interactions. Once this algorithm or others like it are available to the community, it is indeed going to change the focus of what we do as structural biologists, but importantly it’s going to allow us to do more and to focus more on the biological questions than the technology. (How and when it will become available is not entirely clear: John Jumper mentioned “internal discussions” in DeepMind about how to share with the community, and said there would be more news on that next year.) Best wishes, Randy Read > On 4 Dec 2020, at 01:34, James Holton <jmhol...@lbl.gov> wrote: > > It is a major leap forward for structure prediction for sure. A hearty > congratulations to all those teams over all those years. > > The part I don't understand is the accuracy. If we understand what holds > molecules together so well, then why is it that when I refine an X-ray > structure and turn the X-ray weight term down to zero ... the molecule blows > up in my face? > > -James Holton > MAD Scientist > > > On 12/3/2020 3:17 AM, Isabel Garcia-Saez wrote: >> Dear all, >> >> Just commenting that after the stunning performance of AlphaFold that uses >> AI from Google maybe some of us we could dedicate ourselves to the noble art >> of gardening, baking, doing Chinese Calligraphy, enjoying the clouds pass or >> everything together (just in case I have already prepared my subscription to >> Netflix). >> >> https://www.nature.com/articles/d41586-020-03348-4 >> >> Well, I suppose that we still have the structures of complexes (at the >> moment). I am wondering how the labs will have access to this technology in >> the future (would it be for free coming from the company DeepMind - >> Google?). It seems that they have already published some code. Well, >> exciting times. >> >> Cheers, >> >> Isabel >> >> >> Isabel Garcia-Saez PhD >> Institut de Biologie Structurale >> Viral Infection and Cancer Group (VIC)-Cell Division Team >> 71, Avenue des Martyrs >> CS 10090 >> 38044 Grenoble Cedex 9 >> France >> Tel.: 00 33 (0) 457 42 86 15 >> e-mail: isabel.gar...@ibs.fr >> FAX: 00 33 (0) 476 50 18 90 >> http://www.ibs.fr/ >> >> >> To unsubscribe from the CCP4BB list, click the following link: >> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 >> > > > To unsubscribe from the CCP4BB list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 > ----- 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: rj...@cam.ac.uk Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk ######################################################################## To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 This message was issued to members of www.jiscmail.ac.uk/CCP4BB, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
