Two years ago, AlphaFold dominated the free modeling category at CASP13. This year's CASP14 results are out, and AlphaFold has been acknowledged for solving the 50-year-old grand challege of protein structure prediction.
https://deepmind.com/blog/article/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology
Good question, maithra.
This is already being discussed at the CASP meeting that started today:
https://predictioncenter.org/casp14/doc/CASP14_Meeting_Program_1.pdf
The question is what Deep Learning can do for protein design!
David Baker is presenting a "Protein design and covid" talk at this Friday's Covid CASP session.
After the CASP14 conference is over, I would like to see an article from the management re: "What is the future of FoldIt?"
Nature opinion piece on the importance of this result:
https://www.nature.com/articles/d41586-020-03348-4
After a multi-year sabbatical, I didn't come back to FoldIt until April of this year. Did we participate in CASP14? If not, why not? I don't remember reading anything about it here until this post.
The recent CASP targets have tended to be on the large side, and you have to work on each target to compete. Plus those robots are getting better and better.
Here's what beta_helix had to say about it in 2018, after CASP 13: https://fold.it/portal/node/2006272#comment-37895
The way CASP has worked since CASP 11 is that groups must submit models for all targets in a particular category in order to be evaluated.
Since Foldit cannot handle the puzzle load, not to mention the many targets that are over 300 residues, we have focused on protein design and ED... where we did use CASP 13 targets: https://doi.org/10.1371/journal.pbio.3000472.s028
More commentary here:
https://blogs.sciencemag.org/pipeline/archives/2020/11/30/protein-folding-2020
DeepMind cofounder and CEO of Alphabet says he was inspired by Foldit:
https://www.bbc.com/news/technology-55157940
Yep, Seth sent this around to the Foldit team this morning.
We're going to discuss it at our Foldit meeting later today... it merits a news post, don't you think? ;-)
and they had an entire slide about Foldit!
Will post more after the conference ends later today, but wanted to pass this on :-)
I love this passage from Janet Thornton, Director Emeritus of the European Bioinformatics Institute:
"As a lover of everything protein, the most exciting thing for me is that this breakthrough is not an end, but a whole new beginning, bringing with it electrifying opportunities and follow-on questions. The structures allow us to understand better how the proteins function and, in turn, this could enable us to fine-tune this function for the benefit of people and the planet. Just like the Human Genome Project facilitated the birth of new scientific disciplines, such as genomics, solving the protein structure question could bring about new and exciting fields of research. One thing is for sure, it’s a fine time to be a protein scientist!"
https://www.ebi.ac.uk/about/news/opinion/AlphaFold-protein-structure-prediction
CASP/AlphaFold have released the slides from their talk at the CASP conference - kinda hard to follow without the talking part, but interesting nonetheless:
https://predictioncenter.org/casp14/doc/presentations/2020_12_01_TS_predictor_AlphaFold2.pdf
Also here is a long and thoughtful piece about what this advance means for the field:
https://moalquraishi.wordpress.com/2020/12/08/alphafold2-casp14-it-feels-like-ones-child-has-left-home/
That blog post by Mohammed AlQuraishi is a great read: thanks susume.
Another post by Mohammed AlQuraishi here: mostly concentrating on the gory internal details of how AF2 works.
https://moalquraishi.wordpress.com/2021/07/25/the-alphafold2-method-paper-a-fount-of-good-ideas/#s5
Stay tuned tomorrow for some Foldit news about this :-)
(In French) very good video explaining how Deep Learning works.
https://www.youtube.com/watch?v=trWrEWfhTVg&t=1188s
Note:
-the importance of a large amount of data
-the usefulness of an intermediary algorithm where Human describes the relevant characteristics to look for
-the detrimental effect of false classifications in the database
Neural networks with deep learning are not able to propose names for plants and animals on a photo, like in this Belgian-Netherlands Citizen Science project here:
https://eur.observation.org
(you post a picture of an European species, the system proposes you a species name. When it's >80% probability, you further check then ask for a validation by experts => the database enrich all the time with validated observations - it now only works well for BE-NL species).
A tweet today from Demis Hassabis, CEO of DeepMind:
@demishassabis
Brief update on some exciting progress on #AlphaFold! We’ve been heads down working flat out on our full methods paper (currently under review) with accompanying open source code and on providing broad free access to AlphaFold for the scientific community. More very soon!
Also, a very interesting graphic showing where AlphaFold2, trRosetta, Rosetta, and a menagerie of other protein modeling methods fall on the continuum between big-data-based and biological-knowledge-based approaches can be found in a recent review in Cell Systems at:
https://www.cell.com/cell-systems/fulltext/S2405-4712(21)00203-9
Will the Alphafold ability of predicting folding change what we are doing in fold.it? Shall we stop folding proteins?