AlphaFold: The making of a scientific breakthrough
Вставка
- Опубліковано 29 лис 2020
- The inside story of the DeepMind team of scientists and engineers who created AlphaFold, an AI system that is recognised as a solution to "protein folding", a grand scientific challenge for more than 50 years.
Find out more:
deepmind.com/alphafold
Protein references:
TBP = To be published
1BYI: Sandalova, T., et al. (1999) Structure of dethiobiotin synthetase at 0.97 A resolution. Acta Crystallographica Section D 55: 610-624.
3NPD: Das, D. et al. (2014) Crystal structure of a putative quorum sensing-regulated protein (PA3611) from the Pseudomonas-specific DUF4146 family. Proteins 82: 1086-1092.
5AOZ: Bule, P., et al. Structural Characterization of the Third Cohesin from Ruminococcus Flavefaciens Scaffoldin Protein, Scab. (TBP)
5ERE: Joachimiak, A. A novel extracellular ligand receptor. (TBP)
5L8E: Dharadhar, S., et al. (2016) A conserved two-step binding for the UAF1 regulator to the USP12 deubiquitinating enzyme. Journal of Structural Biology 196: 437-447.
5M20: Liauw, P., et al. Structure of Thermosynechococcus elongatus Psb32 fused to sfGFP. (TBP)
5W9F: Buchko, G.W., et al. (2018) Cytosolic expression, solution structures, and molecular dynamics simulation of genetically encodable disulfide-rich de novo designed peptides. Protein Science 27: 1611-1623.
6BTC: Mir-Sanchis, I., et al. (2018) Crystal Structure of an Unusual Single-Stranded DNA-Binding Protein Encoded by Staphylococcal Cassette Chromosome Elements. Structure 26: 1144.
6CL6: Buth, S.A., et al. (2018) Structure and Analysis of R1 and R2 Pyocin Receptor-Binding Fibers. Viruses 10.
6CP9: Gucinski, G.C., et al. (2019) Convergent Evolution of the Barnase/EndoU/Colicin/RelE (BECR) Fold in Antibacterial tRNase Toxins. Structure 27: 1660.
6CVZ: Loppnau, P., et al. Crystal structure of the WD40-repeat of RFWD3. (TBP)
6D2V: Clinger, J.A., et al. Structure and Function of Terfestatin Biosynthesis Enzymes TerB and TerC. (TBP)
6E4B: Tan, K., et al. The crystal structure of a putative alpha-ribazole-5'-P phosphatase from Escherichia coli str. K-12 substr. MG1655 (CASP target). (TBP)
6EK4: Brauning, B., et al. (2018) Structure and mechanism of the two-component alpha-helical pore-forming toxin YaxAB. Nature Communications 9: 1806-1806.
6F45: Dunne, M., et al. (2018) Salmonella Phage S16 Tail Fiber Adhesin Features a Rare Polyglycine Rich Domain for Host Recognition. Structure 26: 1573-1582.e4.
6M9T: Audet, M., et al. (2019) Crystal structure of misoprostol bound to the labor inducer prostaglandin E2receptor. Nature Chemical Biology 15: 11-17.
6MSP: Koepnick, B., et al. (2019) De novo protein design by citizen scientists. Nature 570: 390-394.
6N64: Birkinshaw, R.W., et al. Structure of SMCHD1 hinge domain. (TBP)
6N9Y: Kerviel, A., et al. (2019) Atomic structure of the translation regulatory protein NS1 of bluetongue virus. Nature Microbiology 4: 837-845.
6ORI: Spiegelman, L., et al. Enterococcal surface protein, partial N-terminal region (CASP target). (TBP)
6PX4: Krieger, I.V., et al. (2020) The Structural Basis of T4 Phage Lysis Control: DNA as the Signal for Lysis Inhibition. Journal of Molecular Biology 432: 4623-4636.
6QVM: Osipov, E.M., et al. Crystal structure of native O-glycosylated multiheme cytochrome cf with S-layer binding domain. (TBP)
6T1Z: Debruycker, V., et al. (2020) An embedded lipid in the multidrug transporter LmrP suggests a mechanism for polyspecificity. Nature Structural & Molecular Biology 27: 829-835.
6TRI: Rasmussen, K.K., et al. (2020) Revealing the mechanism of repressor inactivation during switching of a temperate bacteriophage. PNAS 117: 20576-20585.
6U7L: Minasov, G., et al. 2.75 Angstrom Crystal Structure of Galactarate Dehydratase from Escherichia coli. (TBP)
6UBL: Kosgei, A.J., et al. Structure of DynF from the Dynemicin Biosynthesis Pathway of Micromonospora chersina. (TBP)
6UK5: Alvarado, S.K., et al. Structure of SAM bound CalS10, an amino pentose methyltransferase from Micromonospora echinaspora involved in calicheamicin biosynthesis. (TBP)
6VR4: Leiman, P.G., et al. Virion-packaged DNA-dependent RNA polymerase of crAss-like phage phi14:2 (CASP target). (TBP)
6X6O: Shi, K., et al. (2020) Crystal structure of bacteriophage T4 Spackle as determined by native SAD phasing. Acta Crystallographica Section D 76: 899-904.
6XBD: Coudray, N., et al. Structure of MlaFEDB lipid transporter reveals an ABC exporter fold and two bound phospholipids. (TBP)
6YA2: Bahat, Y., et al. First structure of a glycoprotein from enveloped plant virus. (TBP)
6YFN: Rumnieks, J., et al. Expansion of the structural diversity of single-stranded RNA bacteriophages. (TBP)
6YJ1: Sobieraj, A., et al. (CASP target) Crystal structure of the M23 peptidase domain of Staphylococcal phage 2638A endolysin. (TBP)
7JTL: Flower, T.G., et al. (2020) Structure of SARS-CoV-2 ORF8, a rapidly evolving coronavirus protein implicated in immune evasion. Biorxiv. - Наука та технологія
Request : Please also consider making an 30-40 min version documentary of this with some more details of protein folding and nuances. I have watched AlphaGo documentary over and over again(i'm sure many others would have) , and still gives me goosebumps.
Totally agree on this! Really would support both financially and with any other possible efforts and resources the documentary like that. This is a unique topic to be covered!
I actually thought this was a teaser for something of this kind and was bitterly disappointed! The AlphaGo documentary is so damn good
I second this. Even willing to chip in if you are going crowd-funding.
Me too
Seconded!!
I wish this was longer with all the details!
I’m hoping they do a full documentary like they did with alpha go
the paper isnt released yet, so they are prob waiting for that to happen
@@1231legomaniac Same!
I wish it had any details. It began with telling us that protein are important and never exceeded that low level of information.
as a molecular biology major this is sick
4:04 coder starts running... NOW IT'S GETTING SERIOUS 😱😱😱
britches on fire. must find water
He's gonna merge to master! After him!
Bad food.
"The only thing I find more amazing than the rate of progress in AI is the rate in which we get accustomed to it" - Ilya Sutskever
Another aspect which to me, is worrying. If we can not see the incoming train as a threat how to avoid any dangers concerning it?
not a good quote
@@lutZnrw Not a good criticism ;) Tell us why and we can discuss it
"the AI effect", look it up.
"AI is technically an expantion of our capability to process and order information. So it would be unlikely that AI was unfamiliar to us."
- Me
(sorry forgot to mark the quote)
As a protein chemist and structural biologist, I truly am impressed and congratulations to the DeepMind / Alpha Fold team. This is a major advancement and if you are taking requests, please present a solution structure for apolipoproteins. Their lipid-free state has defied analysis for over 50 years and it would be terrific to have a validated base structure to build upon.
I don't understand the distinction between AlphaFold and Rosetta. I'm not a structural biologist but hasn't Rosetta already been doing computational protein structural prediction for years?
@@serenolopez-darwin1975 Structural prediction software like Rosetta have been instrumental in identifying structural domains and conformational states that were not anticipated and led to re-evaluation of data and a more clear idea of the shapes proteins may be adopting and how they may be doing their jobs. But no insult intended to the Rosetta software team but the algorithm(s) used don't necessarily result in a model or set of models that conform to reality. Often the more likely answer is what shape a molecule can adopt, not what conformation it does adopt. What Alphafold does is increases the "accuracy" of the estimated shape tremendously, although I predict that proteins with multiple conformational states or that are highly adaptive will still be elusive targets. A good example are lipoproteins.
@@neousagi Yes I assume that even with google's computing power, IDPs are just too intense to simulate unless MD gets really accurate really fast. Rosetta seems to work fairly well for de novo design though, if only of super-stable proteins. I know it has the issue of trending towards extreme energy favorability while natural proteins don't always, does AlphaFold predict less stable crystal structures accurately?
@@serenolopez-darwin1975 To address your question, first there are three basic areas of protein structural analysis: X-ray crystallographers, computational predictive analysts, and wet lab empirical analysts. Consistently these approaches yield different results for proteins with multiple or dynamic structures. The primary reason is that x-ray crystallography identifies the ground state or unit structure that is most prevalent near ground state. Software algorithms are designed to reproduce x-ray structures (as reported here for AlphaFold) and empirical analysis generally evaluates the native fold, which is not necessarily ground state. Viral fusion proteins are another good example of molecules whose native fold is not the same as their ground state fold. So often empirical data will not match X-ray or predictive analysis and therein lies a core debate in the field. Because X-ray and predictive analysis can display a protein's structure down to atomic level resolution in beautiful ribbon structures, they present convincing impressions of how molecules might fold. So the breakthrough with AlphaFold is not so much that they solved all structures but that they have created ground state prediction that is accurate enough to match crystallographic data. To my knowledge, this does not solve the gap between X-Ray crystallography and empirical data analysis (chief amongst them are the NMR, EPR, and spectrographic scientists).
@@neousagi Ah, I've revealed the depth of my ignorance here, I didn't realize that crystal structures are also always ground state (I'm a computational scientist specialized in genomics and so am entirely unfamiliar with the actual experimental methods of XRC). I always heard that conformation to crystal structure was the gold standard of structural prediction but didn't realize the high-stability problem extended even beyond computational analysis to XRC as well. Thanks for the clarification, maybe now I won't seem so dumb when I talk to the proteomics folks lol
Absolutely huge... props to developers. Look forward to history in the making.
Huge? That’s what she said
I have been waiting on the next bombshell from deepmind and oooooh boy did they ever deliver!! Being on the life science field myself I can appreciate how absolutely monumental this achievement is
Please, elaborate your answer as to possible utilization scenarios. If I may say it this way, if we have the biomedicine equivalent of a 3d printer and right materials, does this mean we can synthesize combative medicine on a need basis?
04:01 Demis: "We need to double down and go as fast as possible from here"
04:04 Pushmeet: sprinting through the lobby, here I come bossss
Jokes aside
i question your being in the life science field @DeezMaster seeing what youtube channels you've subscribed to
@@mehulgoel5875 Pretty idiotic comment btw. How does the selection of people he watches affect his career? The height of petty.
@@nick111138 Agreed. Still.. his answer is so generic. On topic: would be useful for folding transmembrane proteins, that are hard to crystalize and solve the structure of by means of x-ray crystalography. If this can be used for that.
I'm 21 years old, and this is without a doubt one of the biggest scientific advancements in my lifetime
It is and Im so happy to read a comment acknowledging this!
I know right..
Yesterday GPT-3, now its AlphaFold2
KEEP YOUR FOOT ON THE PEDAL
In everyones. AI is the next big industrial revolution. First came fossil fuels, and steam engine, mass production, computing, automatization, internet, then ai
@Billy Bob If you don't see any breakthrough here I highly recommend to read up on the impact of knowing protein structure and then to look into the CASP14 results of AlphaFold.
Please make the background music louder, I can still kind of hear what the people are saying.
You would think that Google has the budget for good video editing, but I guess they spent it all on AI... :P
Is this sarcasm?
Maybe you have some high frequency hearing loss.
@@AgingsAProblemFPS No sheldon, it's not.
@@joannot6706 hey man, let the AI learn
you know we all want a 48min to 1h30 documentary with slow motions, drone shots and epic music
No mate, I just want to know what they are talking about.
@@tensevo 1- Go watch the AlphaGo documentary. 2- Educate yourself on Google's "Deepmind" department 3- Come back to watch this. You'll have a bit more context.
Deepmind: Humanity, we discovered how the proteins fold up!
Humanity: 240K views
.
pull, vibrate...virbrate.. bend..bend.. automode..hmm..ok, it's spaghetti . .
@@8joeb219
i installed foldit a few times
also with the corona viral protein code
but.. it was .. limited
partly locked .. why?
@@taiebshahalizade9115 I agree, this company has clearly worked tirelessly over the past year to establish this revolutionary software. Making it open source straight away is a bit of a kick in the nuts, let them get their due share.
@@8joeb219 Yeah, Imagine pushing this to github
@@8joeb219 I imagine they're still refining it and once they're satisfied with its reliability and accuracy they'll commercialize it in some way. I doubt they'll release it for free after dumping so many resources into it. And if I'm being honest, I think they would be completely justified in charging a fee
Congratulations and thank you to the DeepMind team for pushing science forward!
I love great scientific achievements. The few people who recognize them in this world applauded your life long efforts!
:-D That comment makes you even more stupid than the people in this movie!
@@0Amaretto0 I bet you wouldn't say that to their face now, would you, coward?
@@user-ol5bj4dm2v Hm? Why wouldnt i? Its the truth. I am known for spitting the truth.
Congratulations to the Deep Mind team for solving this previously unresolved problem! Fantastic work.
I wish they could try to improve alpha zero-chess so it can defeat stockfish 16-chess in a 1000 game match
A paradigm shifting moment in our lifetime. Congratulations to all DeepMind team!
This video gave me goosebumps. Kudos to amazing work, documentation of the journey and music. I wish they make a full documentary like AlphaGo.
What a triumph! Congratulations to everyone involved, and humanity thanks you for your efforts.
Congratulations! Your achievement really brought me to tears of deep joy!
Fantastic and inspiring work. There are handful of people who are genuinely changing the world for the better and I believe you guys are one of them!
Indeed, "What will be the next surprise?" is the chief motivator of life. Curiosity and discovery. Excellent work.
Thank you for your service guys. The world appreciates it
I haven't a clue what these geeks are talking about but the music sounds like they've are solving a massive problem. Good luck geeks.
Good luck pac man
These are nerds, not geeks.
Congrats to the team for doing really complex and difficult work.
Incredible work, congratulations! Hoping for a full documentation like AlphaGo!
04:01 Demis: "We need to double down and go as fast as possible from here"
04:04 Pushmeet: sprinting through the lobby, here I come bossss
Jokes aside
FOLDING AT HOME TEAMS HAVE LEFT CHAT
Maravilha. Estou amando poder acompanhar a pesquisa de vocês. Muito obrigada! 😀👍🏻
Congratulation ! Looking forward to a longer video/blog/talk with some more details for us non casuals. ;) Thanks !
Congratulations. I can imagine the feeling of success after 14 years. All the best
This is both incredibly amazing and unnerving. Deep learning has now demonstrated specialized super intelligence beyond rudimentary game theory... The box is being opened and who knows what applications will be developed/ are being developed outside of the public eye.
You guys are so amazing and inspiring, thank you so much for the work you do for humanity. I really hope I can use AlphaFold to model different maternal transcription factors binding together on a segment of DNA for my research.
This is absolutely incredible, well done to everyone involved!!
Congratulations to everyone in the team, that's a significant milestone
I don't know why but i feel so happy for them.
Hop - Deep Learning
Skip - Nanotechnology
Jump - Singularity
then we will find another problem to bound after. unless the singularity kills us.
But all of these techniques are symbiotic : you can’t have quantum computers without jumps in nano fabrication techniques etc
@@philosophyman
Singularity doesn't mean we have no challenges left.
But it does mean we can have more and better choices.
let's goooooooooooooooooooooooooooooooooooo
Congratulations to the team! Great work!
Outstanding! Excited about the future of treatments I can offer to my patients
Never has a group of people looked more like how I'd imagined they would, if I'd only been given their job descriptions.
LMAO so trueee
I hope they are not incels
@@sapitron some of them might sadly be
@@CastaneaMa Even if they are they shouldn't be grouped together with people like twitch mods because they're actually geniuses who are contributing much to humanity.
So what??? You should be very grateful that such people exist...so we all can have a better life!
Those crescendo cellos in the background hit me. Almost shed a tear. Then the Email response happend at 6:50
I studied Biochemistry in 2000, and macromolecules structure was one of the hardest subjects. Glad for this breakthrough finally arrived and looking forward to seeing real-life applicaitons and industry applied technologies soon.
Awsome! Deepmind, you did a really incredible job ever! I respect you, and look forward the next steps of you!
FOLDING AT HOME TEAMS HAVE LEFT CHAT
Folding at home will work even better now, they will hopefully create an AlphaFold package that you can run on your gpu for increased accuracy and speed compared to traditional methods.
they also likely help make the training data for this Ai's formula is basically (training data) + GPU = Magic.
@@default2826 hopefully!
@@default2826 that would be amazing.
@@default2826 If the price drops sufficiently, the complexity of Folding@home may not be worth it.
as someone who has a deficiency which causes protein misfolding in the liver, i greatly appreciate contributions in science like this. and hope there will be more studies like this in the future to prevent diseases...unfortunately things like treatments and "cures" are really expensive but these are still incredible strides anyway!
Thanks simply amazing and thank you, all of you!
it's amazing how many consistent breakthroughs deepmind is making!
How exactly does Google brain fit into all of this. What do they do
Based off their actual data the folding problem isn't solved yet but they are damn close. If they keep on finding tricks (like they did with epistatitic data) to better teach the ML algorithms they will have this solved within a couple years. Keep up the good work.
make sense
Since then, a number of other protein-folding programs have made their debut. They are all faster. AlphaFold 2.3.2 is currently the most up-to-date.
Congratulations to the whole team at DeepMind. This is an incredible achievement. If you could also create an ai to predict the interactions between different structures, that would open up even more possibilities for custom protein design.
Archievment??? :-D IS that a joke? They made up a an idea and needed 50 years to solve their OWN IDEA!!! This couldnt be more funny stupidity... :-D
@@0Amaretto0 what have you achieved in your life little man? What knowledge did you give to humankind?
@@0Amaretto0 you know you’re being a piece of hot fuming garbage who thinks he’s a genius.
@@aedaldaniel I AM life. And i give the highest knowledge that could ever be recieved. But what has it to do with the topic?
@@0Amaretto0 that explains it your knowledge can be received and understood by worms because you produce worm knowledge. Congrats 🎉
This is mind-blowing, inspiring! What a time to be alive.
Congratulations. You solved the static problem. Now we need to find out how the proteins work and move, the dynamics.
"the DeepMind team presents its approach on 1 December"
Where did you find this o.o
@@zeronothinghere9334 From the nature article
It is today where is it !!! Hype
www.nature.com/articles/d41586-020-03348-4
i think I've seen you on cf?!
What a time to be alive!
*two minute paper intensifies*
Congratulations. Absolutely amazing achievement.
This is epic!! I'm very excited to see what becomes possible with this technology.
I use AlphaFold for my biochemistry research! Amazing stuff, I used it to predict whether a mutation on a specific amino acid site would be possible on a beta-glucosidase enzyme and after weeks of work - the mutation actually expressed!
Really?? How did you get access to alphafold? I want to deduce a structure, where do I submit?
@DeepMind are you going to publish a second video to give a more detailed explanation of how the model is implemented and what kind of improvement did you implement to solve the protein fold problem?
The still have yet to publish their paper on it, give them some time I guess
@@Deez-Master, I am fully aware of that. I just wanted to emphasize that it would be great to see a proper paper review given by the authors (it would be like Einstein explaining general relativity). Obviously, I can do it by myself or wait for some UA-camrs to prepare one. Still, it would be much more interesting to see the authors describing their project emphasizing all the moments when they struggled and how they figured out the solution.
@Adam Dendek
ua-cam.com/video/B9PL__gVxLI/v-deo.html
this is from Yannic Kilcher, he explains paper and many research in this area. Who explained quite in details the abstract from Alpha Fold.
Fun fact one of the woman engineer u saw in the video, was an intern under him :P
I think it is a graph nn based transformer
@@asifdomo500 fun fact: "woman engineers" are called engineers
As an aspiring Biochemist I am excited to see the final result!!!!!.... It just makes science even more exciting
I was weeping while I heard the news. This IS a game changer!
When we all work together amazing things happen
Imagine they can make 2D Waifu based on Virtual Protein.
The important questions in life
I need 10's of these
Maybe in 50 or 60 years in the future.
That's great work... congratulations Team AlphaFold.
Amazing stuff. Just starting out in the field of structural biology this year.
Remember learning about protein structures in school and asking exactly if we couldn't just figure it out from the base sequence since it was deterministic. Of course my teacher told me that would never be possible due to the complexity. Glad "never" just means a few decades in the scientific fields
The machine of the world runs to power endeavours such as this. I hope you all know how important you are to humanity as a whole.
You are so inspiring. Excellent work.
Hats off again from a long time StarCraft fan. This time you’re saving the world!
Congratulations to the whole team,
Shoutout to Sardarji,
Love and respects from India.
1:03 That Marriages band shirt!🤘
Superb! Sharing this here and there...
the level of these animations is outstanding
Towards the end of this event, the competing teams just completely folded.
you could say they were pro team folding. I'll get my coat.
@@ClayMann That kind of humor works for me. :-)
i guess you've got to know when to hold em, and when to fold em
Really nicely made piece covering an extremely complex subject.
Wow, this is an amazing achievement !!
I swear this group of people emits intelligence.
I would give everything to spend one week with them and just listen and observe.
Awe inspiring! Thank you for working so hard for man kind. A little more fuel to finish writing my PhD...
Congratulations DeepMind! A staggering achievement :)
Thank you for your efforts👏
The "next surprise" is the Paperclip Accumulator scenario.
Great job 👏🏼👏🏼👏🏼👏🏼 and congratulations to the Deep Mind team! Could y’all please please please make a full-length documentary feature on this epic saga like you did with AlphaGo? That would be bloody awesome! People please like this comment if you agree! 🙏🏼
Damn you guys! You did it again! Congratulations!
It's ALL about the chart at 7:26. Amazing!
I hope y'all be joining my "How to make Simon rich af" science competition this year?
@Grandfather_Din_Racket Yeah i am quite a stochsatic benchmark but that‘s where the fun starts
This is actually the true story of how Google was founded
an open source AI that is a stock traders best friend is right around the corner (within 2 years)
Making DeepMind open-source will be revolutionary for us biologists/biotechnologists.. so please do so.
They surely dumped millions and millions of $$$, time and dedication into this. You can't expect them to just give it away for free. This is not how the world works and they should reap the fruits of their labor.
@@StevenKger I'm sorry. In that case I hope that everybody profits from it as well as possible.
Amazing achievement, congratulations!
Amazing stuff, we know when Demis say we have gone AGI we are in a knew world in the making.
Damn, this made me emotional and proud for our kind...
Structural biologists, computational biologists or computer scientists?
@@subhrodeepsaha9245 Humankind.
@@Shadowbladekokotil Was so caught up in the science that I didn't realize something so obviouse 😅😅😅
How stupid of me 😂
Please, a round of applause for the 46 people that believe coronavirus isn't real. Great work, PLEASE keep working on this😍
What a great team. What a great project.
Why does this give me goosebumps?
Hyy Deep Mind.. Great going
But just wanted to let you know that ur logo looks like Doordarshan 😆😍
Waiting for the chinese version of CASP14,
Because it would be a sign that the Chinese gov and academia are pursuing AGI too
Such very important and significant work.
The way the video is spliced together makes everything rather exciting; running around, standing in circles for group meetings, etc. In this group of seemingly very clever people, I find it remarkable to see 26-27 years of work crunched into 8 minutes.
I also still have very little idea what CASP is and how AI relates to protein folding.
@@TheComicChild by statically manipulating we can predict through patterns, how they move, why they move etc.
The fact that this video doesn't have 7 billion views is the reason why I cry myself to sleep at night.
Me in CptS 121 writes an algorithm to reverse a string: This
This was big news. Amazing work!
Congratulations for making history !!!
Hi there, I work at the MailOnline, can we use your video in our player with a credit back to you? many thanks, Phil
Hi Philip - sure, usage with credit is fine! Thanks
I think I've discovered enough seeing you wear those crocs at home
this really blows my mind!!! 🤯
well done to everyone. brilliant young minds
That's it i'm quitting my biology degree and going over to software engineering, i've been fascinated with computers for so long