A quick video on the basics of DeepMind's AlphaFold 2 breakthrough. Here are the timestamps: 0:00 - What happened? 1:03 - How big is this accomplishment? 4:39 - Proteins and amino acids 5:17 - Protein folding 8:26 - How AlphaFold 1 works 9:45 - How AlphaFold 2 works 12:09 - Why is this breakthrough important? 13:19 - Long-term future impact Please support this channel by checking out our sponsors: - Vincero: vincerowatches.com/lex to get up to 25% off + free shipping - Four Sigmatic: foursigmatic.com/lex and use code LexPod to get up to 60% off
To the question of, do we like these short videos? Absolutely, and you said that you really enjoy making them. Double positive. Keep up the amazing work Lex
Big respect for you man, been following you since you only had 50k followers. You are such a great mixture of intelligence, wisdom, respect, dry humor, thoughtfulness, open-mindedness, and integrity. Hardly ever do you see a person climb to the highest heights of academia, science, athletics/martial arts, and personal development (and maintain their sense of humility I might add)....Cheers Lex, from a fellow engineer and martial arts enthusiast, I'm proud of you and I can't wait to see what 2021 has in store for you man
And above all, look at his eyes any time he talks about love. He is one of the best humans alive today. I really wish we could make him a president of the world.
The moment I heard of this breakthrough, I was excited for the sure to come video of you explaining it. And you haven't disappointed me one bit! This was such a well structured and explained video and I love your personality that goes with the explaining (the occasional poetic or philosophical thoughts). Thank you so much for this video!
Please make more of these! I bet a lot of your viewers, like me, are undergrads in compsci. Seeing where our degrees could take us and how the skillset we're learning can be applied in a byte-sized video, that can be watched while on the toilet in the morning is great.
You inspire me in many ways lex. You alone have spawned a massive interest in deep learning and AI in me and I can’t thank you enough. I just wanted to chime in just in case you see this comment and let you know you continue to massively positively influence my life. I really appreciate the time and love you put in to the things you release to the internet and I do my best to reciprocate that in my own ways every day.
I've started to despise his overly romantic side. This is distracting me and turning me off. I've started watching less of his content of late. Others like it, so it's likely a win
@ozafter1 So becoming a "real" adult has nothing to do with extreme understanding of themselves and the world around them, but better yet having kids and developing a jaded world view? Maybe I'm being unfair to your argument but life with or without kids or a wife is not any less of a life. Lex talks about heavily impactful things, life is weird, and no one likes a bore. He asks questions and entertains concepts that most people's egos would be far too afraid of even bringing up, no matter how trivial the sentiment itself might actually be. We see this a lot with Rogan too. This is why people enjoy their content to the extent they do. Rogan has kids and a wife by the way.
@@nateo7045 i didn't say that. i don't think he has an extremely deep understanding of himself (except in terms of learning and discipline), other people, or society at large. he's comically naive on human relationships and business. but whatever, i love the guy and this podcast is nothing short of amazing to me.
Keep doing these Lex, they are so well done and explain a ton of good info in a reasonable amount of time! I love how the short term and long term implications are broken down nicely. I love the slide deck design too.
Halfway through the video... so far I have not understood one word of what is going on. Kind of happy about that and VERY glad that there are people like you around that can keep an eye on this stuff.
@@tn15_ Ya it's too bad he could only get people like: Andrew Ng, Elon Musk, Stephen Wolfram, James Gosling, Francois Chollet, Richard Karp, Jack Dorsey, David Silver (who btw is a lead researcher at deepmind), Grant Sanderson, Daphne Koller, Donald Knuth, Noam Chomsky, Richard Dawkins, Garry Kasporav, Bjarne Stroustrup, Daniel Kahneman, Leonard Suskind, Jeff Atwood, Yoshua Bengio, Gilbert Strang, Michael l. Jordan, Richard Feynman, Alan Turing, Albert Einstein, and other "low tier" guests to come on. But in all seriousness, the quality of this podcast is really superb; the fact that he interviewed everyone on that list besides for the last three is really amazing.
@@neslef3 I meant no slight to the quality of the podcast with my comment -- only that I was interested in hearing more from DeepMind and I figured Lex had the pull to get the interview. I only stumbled upon this channel 4 months ago, so I still have a lot of catching up to do! Also, I agree with you. The content on here is phenomenal and unlike much else on YT.
@@tn15_ Definitely didn't think you were saying anything bad about Lex, if anything thought maybe you just didn't know about him. I mostly just wanted to take the moment to list some of the incredible people he had on the podcast as he never ceases to amaze me with his guests!
I just spotted this video. Lex! Yes, by all means, please keep on creating these educational lecture segments. You have a terrific way of breaking these complex lessons down to be understandable and fun to analyze.
Beyond the monotone voice you can hear his enthusiasm. Just hearing how excited he is to talk about this makes me excited to hear about it. Love beyond bounds for Lex!
For a layperson: What is the next step now that the problem is "solved" ? What is "the thing that really really needs to not have an unneccesary bottleneck now" ?
Ah ein Stein: If they've truly solved this problem, don't expect it to come cheap. deepmind was working with the uk national power grid to optimize power distrubution but, someone that wants to stay anonamous, from the uk power grid said this: "The money DeepMind was asking was outrageous,” the source said. “Most of their work is internal and they bill Google for it,” the source added. “They sell the work of their AI engineers at super inflated prices, and not at the price that the market values their output.” Needless to say, a deal was never reached.
@@cyberneticbutterfly8506 The issue with all of this is - as was the case with crystal structures - that they show just what the name says; the situation in the crystal. Proteins are not a static thing. To perform their function they have to change in shape (extreme example DNA Pol opens and closes in a 40° angle between domains). So the next step is goin from crystal to solution. I'm not an expert on the computational side of things but I'd guess we're quite far off on that. Don't get me wrong, this is amazing and incredibly useful, but the next step is just as hard.
@@Greg.92 So what you're saying is alphafold's breakthrough is only able to predict protein structure but not protein function? Can they now design new proteins or can they only predict structures of already existing proteins?
My dad is a molecular biologist and runs a biochem company that specializes in enzyme technologies. His work revolves around creating proteins(enzymes) that are high performant and can massively increase production efficiency for drugs/ingredients and etc. This means a huge amount of laborious lab testing the performance of many proteins they created from the ground up. Computational prediction of protein structures was a frequent topic of our casual discussions, with him often being pessimistic about it, saying that proteins are the most complex structures in the universe, and the combinations are infinitely endless, and that only a quantum computer can begin to qualify such a task. Because I work in I.T, I tend to console him with the notion that A.I algorithms these days are surprisingly powerful, and maybe more suitable for such endeavors. We've had this type of dialogue for years. A few days ago when I sent him the news of this breakthrough he was in complete shock. He said this will change the whole game for the entire drugs and biotechnology industry.
Holy Mother-of-all-A.I. developments, Batman! This is bigger than velcro. I mean, even for those of us with very limited cognitive skills can see this discovery certainly changes almost everything in life sciences. This was another superb video, Mr. Fridman, bolshoy spasibo ^10.
This I is what I was waiting for ❤️ thanks Lex , you are amazing. Waiting for your next video with self driving car model test and you playing a guitar. How do you manage all these things at same time is amazing. Keep rocking and keep hustling.🔥🔥🔥🔥
Thank you for highlighting this breakthrough Lex! I would not have realized that something so incredible had been discovered and how much potential a discovery like this holds unless an expert like you had called attention to it, thank you!
Damn, what a time to be alive! Also, huge thanks to you Lex, for all the work you do and such videos especially. I was just thinking about learning more about AlphaFold and protein folding in general when I read the news.
Great Job! I particularly liked when you nerded-out at the end with the mind blowing awe-ness of encoded systems of proteins. Just like mathematics, computer language, and esoteric studies, Biology contains encoded information in plain sight. That 4 simple molecules in our DNA control all the diversity in Life as we see it and may be seemingly linked through gradual evolution, or even the daily process in an organism from their DNA through to every protien-fxn they need in a day/life/life cycle is computationally mind-blowing and YES ALPHA-FOLD WILL OPEN THE DOOR TO A NEW FUTURE. Whew, See i just nerded-out too.
Thanks for the breakdown, Lex. I've been excited to read about what is coming out of the AlphaFold news but your breakdown, as usual, is the most entertaining and informative.
White Ness Not really sure what you're trying to say. Seems like there was a good breakthrough and I was excited to learn about it. What do you want me to doubt? Doubt that the problem is solved? I don't have the expertise for that. Let's hear what those experts have to say. Doubt that the solution will fix anything? Again, don't have the expertise for that, but it looks promising. If you have doubt, then get to work and let us know your perspective.
I am a Canadian Masters Student in Computer Science and with a Bachelor Honors in Biochemistry. I actually just wrote a research proposal for a grant that was about the functional prediction from protein structure. So I have recently read up on this literature a bit. Not saying I am an expert though! I think the biggest implications are for protein engineering. Since proteins (Enzymes to be specific) are responsible for the metabolites of all living organisms, being able to design enzymes to produce natural products efficiently is incredible for the production of drugs and almost any other material tbh. Currently, there is literature on how to predict optimal binding sites, but turning that into a functional enzyme is tough. This takes lots of expensive iterations. This is because our prediction models are bad and predicting if proteins are stable to fold. Because models are trained on natural proteins that function the programs aren't great at telling us what is stable or not, just what a possible fold maybe. but with advancements like these, we will likely get closer to predicting the stability and function of proposed sequences making engineering a possibility. Another huge implication is the possible discovery of uncultivatable proteins. A significant amount of biomass on earth cannot be grown in a lab. This leaves an ungodly amount of organisms and their components unstudied. But, the field of metagenomics takes DNA and RNA samples found from soil or other samples and reconstructs potential genomes. From genomes, we can predict possible amino acid sequences and from that, we can predict natural proteins that we would have never seen before!!! There is always a risk that since the data is from organisms that are cultivatable that the proteins of others may be different in some way, but that is extremely unlikely to be an issue in my opinion. Very Exciting!! On top of this, if we can predict ligand/substrate binding then we can see natural biological compounds that we haven't seen in the lab before! Natural products have a high efficacy in drug discovery so this is fantastic! Big things that still need to be done, in the field are stability prediction, function prediction, ligand/substrate binding prediction, product prediction, dynamic movement prediction for enzymes that go through conformational changes. Lots ahead, but breakthroughs like this make me super hopeful! You're an inspiration Lex! I am hoping to make an impact on substrate/product prediction myself with applications of deep learning!
Really impressive work by Lex in just a few days. I get a lot out of this and hope to see more regarding this breakthrough. Particularly I realized that I need to learn more about transformers. I'm not sure what it means but seems similar to the idea that has been on my mind for a while regarding the determination of relevant features...
Mind was blown when I first read this. I think a lot of people don't quite understand just how amazing this is (or at least will be in the near future). I'm not in drug design etc for now, but I work in medicine. This is completely world-changing for so many entities out there depending on crystallography or just trial and error.
Researcers from DeepMind have the greatest contribution to this major achievement by finding the right approach for such a complex process to be sorted out
Great video!! Please keep doing these type of video's. As an AI student at a university in the Netherlands it's great to keep up to date with reality in this way, thank you.
CRISPR-Cas9 was huge, and we all knew it’d win the Nobel in chemistry or medicine at some point. This is just as big, and could win chem, medicine or physics. If your work is mentioned in three Nobel fields, I can’t even find words for how amazing what you’re doing is. This will change our lives forever even if we don’t notice.
Just one thing about misfolding. Misfolding are of not too big interest to most molecular biologists. It does play a role in some diseases, but most of diseases are caused by other things. Biologists just want to know the structure to know how it interacts with other molecular machinery, be it healthy or unhealthy parts, or in pathogens. The 3D structure is just useful for many things, be it progressing knowledge, drug design, etc. Also a lot of proteins are composed of more than just amino acids. They can have so called post-translation modifications (happens after translation or during), that do include extra attached groups, be it non-amino acids, or short peptide groups, or sometimes non-peptide groups (called cofactors or prostethic groups), thus forming a bit branching strcture (kind of tree in CS), or be composed of multiple sub-units of the same or different type, that together create some weird complexes. All of them determine 3D structure, stability, folding and function. So the complexity is actually way higher than one would thing just by looking at the sequence. I don't belive sequence itself fully describe the protein. It does in many cases, but not always. Anyhow, biologists just want to know the structure of correctly folded protein for many reasons. Study of misfolding is rather niche, but of course important to some stuff, and progressing knolwedge of the folding process itself. Also I think the 10^143 figure, is just deceptive. It is not like nature tries all the combinations. Proteins are mosly synthetized in sequential process, one amino acid at the time, not in parallel, so obviously, the folding process is mostly incremental, and determined or rather influenced by the enzymes that do the synthesis - folding happens during synthesis naturally. It is like saying air molecules in the room I am in can be in one of 10^100000000000000000000000000000000 possible configurations, but it always finds the energy minimum and spreads uniformly. Or that there might be 10^1000000 books possible, yet humans, always write ones that do have a structure and sense and can be parsed. It silly to even mention this paradox, because it is not a paradox.
Thanks for the video Lex! I couldn't be more excited about this breakthrough. The profound implications for the future (years, not decades) of health really hit me in a deep way and make the AI revolution feel like it's well and truly underway.
I'm doing PhD study in Computational Chemistry and I'm using ML code to research new area of materials. I haven't got results yet but it performs extremely better in most of side than conventional modelling method so far.
This is a great video with a useful insight into how DeepMind came to win CASP. As a biochemist and bioinformatician who has been part of the effort to "solve the protein folding" problem I would say that strictly speaking the problem has not actually been solved. I say this because the research many people focused on was "how do proteins fold into specific configurations?" The DeepMind solution does NOT answer that question. There is no insight into the biophysical rules that guide the folding. And the reason that research has centred specifically on understanding "how" is because the answer to THAT question gives answers in BOTH directions of the problem. DeepMind can go from sequence to structure, but it cannot go from structure to sequence ("I want a protein that has this structure - what sequence gives me that?"). Of course, with further AI and computing power, it is possible to start from a structure that has a similar form to what you wish to create and therefore it is just a matter of applying a genetic algorithm which does what it was created to mimic, to go through the combinations until the target structure is reached. But still, it isn't "understanding", it is just doing. The other thing I would say that DeepMind does not do is offer insight into diseases caused by misfolded proteins. And to be specific here, a misfolded protein is a protein that has not folded into the configuration that its sequence predicts to fold into, usually because of unusual factors within the folding environment. DeepMind can only go from sequence to structure. It cannot provide any information about the folding process or influences of the folding environment. For diseases where "misfolding" is due to sequence mutations strictly speaking the protein is folding correctly, just not the same as the native sequence. For these mutated sequences, yes, DeepMind will help. I have said for a long time that the problem of protein structure prediction would be broken by the brute force of AI. But the protein folding problem still exists.
this is so fascinating. machine learning, AI, genetics, epigenetics, quantum computers, fusion... my greatest wish is that i live long enough to see many of these future developments
One of the better reports on this breakthrough, thanks Lex. In reality there are really only 20 naturally occurring proteinogenic amino acids since Selenocysteine which is homologous to cysteine replaces a Sulfur atom with Selenium and is only found in about 50 proteins. X-ray crystallography will still be around where there is a need to solve more complex protein structures where subdomains exist or to understand how small molecule drugs bind to protein targets. Maybe that will be the next breakthrough but a much tougher problem since there is not a big training set for AI. It will be interesting to know how well this methodology can be applied to predict variations in protein folding brought about by mutations in protein sequences and other important applications such as designer proteins if that is possible.
A quick video on the basics of DeepMind's AlphaFold 2 breakthrough. Here are the timestamps:
0:00 - What happened?
1:03 - How big is this accomplishment?
4:39 - Proteins and amino acids
5:17 - Protein folding
8:26 - How AlphaFold 1 works
9:45 - How AlphaFold 2 works
12:09 - Why is this breakthrough important?
13:19 - Long-term future impact
Please support this channel by checking out our sponsors:
- Vincero: vincerowatches.com/lex to get up to 25% off + free shipping
- Four Sigmatic: foursigmatic.com/lex and use code LexPod to get up to 60% off
16:00 Quick shout out to sponsors
I was hoping you’d comment on this
You're freaking beautiful, Lex
I am deeply skeptical about the generalizability of this. I hope I am wrong.
@@eliriekeberg7127 Obviously we haven't seen the paper yet, but the nature of competition makes it so that it's very likely this is generalizable.
To the question of, do we like these short videos? Absolutely, and you said that you really enjoy making them. Double positive. Keep up the amazing work Lex
*raises hand*🙋🏼✌🏻😁✅🟢🦾💯😋🌟😎🔥
i second this motion & share these thoughts
It'd be nice to see his face too tho
Thanks Lex, love your stuff.
Big respect for you man, been following you since you only had 50k followers. You are such a great mixture of intelligence, wisdom, respect, dry humor, thoughtfulness, open-mindedness, and integrity. Hardly ever do you see a person climb to the highest heights of academia, science, athletics/martial arts, and personal development (and maintain their sense of humility I might add)....Cheers Lex, from a fellow engineer and martial arts enthusiast, I'm proud of you and I can't wait to see what 2021 has in store for you man
Only reason I’m here to emanate Lex
...what this guy said!
And above all, look at his eyes any time he talks about love. He is one of the best humans alive today. I really wish we could make him a president of the world.
Would love more of these mini-lecture episodes. Great work Lex
Yeah...now all he has to do is crank out a steady stream of these one-in-a-lifetime scientific advances and he'll get right on it!
@metoo :)
I love those short videos that touch on recent advancements in AI - more please. Would love more technical details.
The moment I heard of this breakthrough, I was excited for the sure to come video of you explaining it. And you haven't disappointed me one bit! This was such a well structured and explained video and I love your personality that goes with the explaining (the occasional poetic or philosophical thoughts). Thank you so much for this video!
You would make the world's best professor. Super easy to understanding for dense material. A little dry humour to make sure you're paying attention.
Thanks Lex, I really enjoy this style of video. Please keep them coming!
Please make more of these! I bet a lot of your viewers, like me, are undergrads in compsci. Seeing where our degrees could take us and how the skillset we're learning can be applied in a byte-sized video, that can be watched while on the toilet in the morning is great.
Please keep making quick videos like this talking about topics you find fascinating
@Lex Fridman stfu fake Lex
This is such a huge accomplishment! Great video as always.
these short videos are great! I read the alphafold blog earlier, and now watching your video!
@Lex Fridman Keep trying to scam people, fake Lex.
Yes, please make more vids like this. Appreciate the concise, informative report.
"Hopefully I'm not being too poetic" 😂 that's why we love you Lex
You inspire me in many ways lex. You alone have spawned a massive interest in deep learning and AI in me and I can’t thank you enough. I just wanted to chime in just in case you see this comment and let you know you continue to massively positively influence my life. I really appreciate the time and love you put in to the things you release to the internet and I do my best to reciprocate that in my own ways every day.
I've started to despise his overly romantic side. This is distracting me and turning me off. I've started watching less of his content of late.
Others like it, so it's likely a win
@@DheerajBhaskar I mean he's very young, he's never been married, has no kids, is basically an adult child who's brilliant
@ozafter1 So becoming a "real" adult has nothing to do with extreme understanding of themselves and the world around them, but better yet having kids and developing a jaded world view? Maybe I'm being unfair to your argument but life with or without kids or a wife is not any less of a life.
Lex talks about heavily impactful things, life is weird, and no one likes a bore. He asks questions and entertains concepts that most people's egos would be far too afraid of even bringing up, no matter how trivial the sentiment itself might actually be.
We see this a lot with Rogan too. This is why people enjoy their content to the extent they do. Rogan has kids and a wife by the way.
@@nateo7045 i didn't say that. i don't think he has an extremely deep understanding of himself (except in terms of learning and discipline), other people, or society at large. he's comically naive on human relationships and business. but whatever, i love the guy and this podcast is nothing short of amazing to me.
Keep doing these Lex, they are so well done and explain a ton of good info in a reasonable amount of time! I love how the short term and long term implications are broken down nicely. I love the slide deck design too.
Lex this is one of the best videos I've seen on your channel! As a current BME student this was immensely helpful to my understanding - thank you!
@Lex Fridman Keep trying to scam people, fake Lex.
Hmm yes, I’ll use this as copy-paste.
Ever since I heard the announcement I was hoping to see you do a video about it. Thank you Lex!
Love these mini-lectures. Your slides are next-level
Halfway through the video... so far I have not understood one word of what is going on. Kind of happy about that and VERY glad that there are people like you around that can keep an eye on this stuff.
Thank you. Wonderful and timely. Your podcast has provided great content, especially during the necessary social distancing of this year.
@Lex Fridman Keep trying to scam people, fake Lex.
Great video ... clear ... great topic ... and the correct message: "try to learn something new every day ..."
*dead serious monotone educational voice*
"I'm just here to have some fun"
I've heard stronger emotional inflection at a Stephen Hawking lecture.
Im just here for the data lol
reminds me of Charlie Rose from the 80s,,and i love it,,its my ASMR
It's LEX .. he is not that monotone .. just microtonal..
@@sudoall i know lol but i am here to have some fun
this is a great summary. thank you
Great video! quick and important content! very inspirational as always.
Keep doing the videos Lex.. you are helping me to learn something new every single day..❤️❤️
@Lex Fridman Keep trying to scam people, fake Lex.
great summary of the new breakthrough. I think more of these shorter videos would do well, I enjoy them
Next podcast idea: Interview with one of the team members of the AlphaFold project.
That would be great if he could convince someone from Deepmind to come on
@@aryangod2003 I didn't know that. Thanks!
@@tn15_ Ya it's too bad he could only get people like:
Andrew Ng,
Elon Musk,
Stephen Wolfram,
James Gosling,
Francois Chollet,
Richard Karp,
Jack Dorsey,
David Silver (who btw is a lead researcher at deepmind),
Grant Sanderson,
Daphne Koller,
Donald Knuth,
Noam Chomsky,
Richard Dawkins,
Garry Kasporav,
Bjarne Stroustrup,
Daniel Kahneman,
Leonard Suskind,
Jeff Atwood,
Yoshua Bengio,
Gilbert Strang,
Michael l. Jordan,
Richard Feynman,
Alan Turing,
Albert Einstein,
and other "low tier" guests to come on.
But in all seriousness, the quality of this podcast is really superb; the fact that he interviewed everyone on that list besides for the last three is really amazing.
@@neslef3 I meant no slight to the quality of the podcast with my comment -- only that I was interested in hearing more from DeepMind and I figured Lex had the pull to get the interview. I only stumbled upon this channel 4 months ago, so I still have a lot of catching up to do! Also, I agree with you. The content on here is phenomenal and unlike much else on YT.
@@tn15_ Definitely didn't think you were saying anything bad about Lex, if anything thought maybe you just didn't know about him. I mostly just wanted to take the moment to list some of the incredible people he had on the podcast as he never ceases to amaze me with his guests!
Please make more of these videos! this format is easy to digest esp
Man what a time to be alive.
Thanks for the summary. Very cool, quick and concise.
Thanks for this break down Lex, super interesting
Lex, don't dare to ever stop making videos like this!
I just spotted this video. Lex! Yes, by all means, please keep on creating these educational lecture segments. You have a terrific way of breaking these complex lessons down to be understandable and fun to analyze.
This was a good roundup. Great new video with Demis too... keep em coming
Thank you for this video. Really nice work. Definitely would love to see more of such kind of short educational videos.
@Lex Fridman Keep trying to scam people, fake Lex.
Beyond the monotone voice you can hear his enthusiasm. Just hearing how excited he is to talk about this makes me excited to hear about it. Love beyond bounds for Lex!
That was a great explanation of protein folding and alphafold2. Please make another video explaining alphafold2 once the paper releases
Great video as usual!
For someone who worked in a field related to xray crystallography hearing this announcement was both mindblowing and eery at the same time.
For a layperson: What is the next step now that the problem is "solved" ?
What is "the thing that really really needs to not have an unneccesary bottleneck now" ?
Do you happen to work with Quasicrystals?
Ah ein Stein: If they've truly solved this problem, don't expect it to come cheap. deepmind was working with the uk national power grid to optimize power distrubution but, someone that wants to stay anonamous, from the uk power grid said this: "The money DeepMind was asking was outrageous,” the source said. “Most of their work is internal and they bill Google for it,” the source added. “They sell the work of their AI engineers at super inflated prices, and not at the price that the market values their output.” Needless to say, a deal was never reached.
@@cyberneticbutterfly8506 The issue with all of this is - as was the case with crystal structures - that they show just what the name says; the situation in the crystal. Proteins are not a static thing. To perform their function they have to change in shape (extreme example DNA Pol opens and closes in a 40° angle between domains). So the next step is goin from crystal to solution. I'm not an expert on the computational side of things but I'd guess we're quite far off on that. Don't get me wrong, this is amazing and incredibly useful, but the next step is just as hard.
@@Greg.92 So what you're saying is alphafold's breakthrough is only able to predict protein structure but not protein function? Can they now design new proteins or can they only predict structures of already existing proteins?
Thank you so much for this video! I'm late to this fascinating and moving party that is your channel. It's growing on me very rapidly, though!
Thanks Lex. Groundbreaking development explained in groundbreaking video. You're honest, passionate and we trust you. Thanks a lot.
My dad is a molecular biologist and runs a biochem company that specializes in enzyme technologies. His work revolves around creating proteins(enzymes) that are high performant and can massively increase production efficiency for drugs/ingredients and etc. This means a huge amount of laborious lab testing the performance of many proteins they created from the ground up. Computational prediction of protein structures was a frequent topic of our casual discussions, with him often being pessimistic about it, saying that proteins are the most complex structures in the universe, and the combinations are infinitely endless, and that only a quantum computer can begin to qualify such a task. Because I work in I.T, I tend to console him with the notion that A.I algorithms these days are surprisingly powerful, and maybe more suitable for such endeavors. We've had this type of dialogue for years. A few days ago when I sent him the news of this breakthrough he was in complete shock. He said this will change the whole game for the entire drugs and biotechnology industry.
Holy Mother-of-all-A.I. developments, Batman! This is bigger than velcro. I mean, even for those of us with very limited cognitive skills can see this discovery certainly changes almost everything in life sciences. This was another superb video, Mr. Fridman, bolshoy spasibo ^10.
Your videos are very very informative and teach concepts that matter(to me at least). Grateful for your time and effort. Keep dishing out more!!
A Lex Fridman video on AlphaFold2 was not what I expected but definitely what I needed. Thank you for everything you do!
glad you got this up so fast!
@Lex Fridman Keep trying to scam people, fake Lex.
This I is what I was waiting for ❤️ thanks Lex , you are amazing. Waiting for your next video with self driving car model test and you playing a guitar.
How do you manage all these things at same time is amazing. Keep rocking and keep hustling.🔥🔥🔥🔥
Thank you for highlighting this breakthrough Lex! I would not have realized that something so incredible had been discovered and how much potential a discovery like this holds unless an expert like you had called attention to it, thank you!
I love the quick video format lex! Perfect to squeeze in before work 😃
@Lex Fridman Keep trying to scam people, fake Lex.
“Inspiring beyond words” that’s right!
This is where lex shines, excellent presentation
Damn, what a time to be alive!
Also, huge thanks to you Lex, for all the work you do and such videos especially. I was just thinking about learning more about AlphaFold and protein folding in general when I read the news.
Great video Lex! Keep them coming.
Great Job! I particularly liked when you nerded-out at the end with the mind blowing awe-ness of encoded systems of proteins. Just like mathematics, computer language, and esoteric studies, Biology contains encoded information in plain sight. That 4 simple molecules in our DNA control all the diversity in Life as we see it and may be seemingly linked through gradual evolution, or even the daily process in an organism from their DNA through to every protien-fxn they need in a day/life/life cycle is computationally mind-blowing and YES ALPHA-FOLD WILL OPEN THE DOOR TO A NEW FUTURE. Whew, See i just nerded-out too.
This is super helpful Lex, please keep these videos up! Thanks!
Amazing accomplishment 👏.
Unbelievable, Bless you all working on this 🙏
This video is a great overview. I hope to see your coverage when the paper comes out. Thank you.
This is one of the most significant advances for mankind!!!
I remember when scientists announced the breakthrough in Cold Fusion. We were pretty excited about it back then. Haven't seen much it since.
That is really inspiring, the progress we see in all subjects of science is far beyond my expectations. Thanks Lex for giving this insight
Thanks for the breakdown, Lex. I've been excited to read about what is coming out of the AlphaFold news but your breakdown, as usual, is the most entertaining and informative.
White Ness Not really sure what you're trying to say. Seems like there was a good breakthrough and I was excited to learn about it. What do you want me to doubt? Doubt that the problem is solved? I don't have the expertise for that. Let's hear what those experts have to say. Doubt that the solution will fix anything? Again, don't have the expertise for that, but it looks promising. If you have doubt, then get to work and let us know your perspective.
Great video Lex, this format is fun and really digestible.
I am a Canadian Masters Student in Computer Science and with a Bachelor Honors in Biochemistry. I actually just wrote a research proposal for a grant that was about the functional prediction from protein structure. So I have recently read up on this literature a bit. Not saying I am an expert though! I think the biggest implications are for protein engineering. Since proteins (Enzymes to be specific) are responsible for the metabolites of all living organisms, being able to design enzymes to produce natural products efficiently is incredible for the production of drugs and almost any other material tbh. Currently, there is literature on how to predict optimal binding sites, but turning that into a functional enzyme is tough. This takes lots of expensive iterations. This is because our prediction models are bad and predicting if proteins are stable to fold. Because models are trained on natural proteins that function the programs aren't great at telling us what is stable or not, just what a possible fold maybe. but with advancements like these, we will likely get closer to predicting the stability and function of proposed sequences making engineering a possibility.
Another huge implication is the possible discovery of uncultivatable proteins. A significant amount of biomass on earth cannot be grown in a lab. This leaves an ungodly amount of organisms and their components unstudied. But, the field of metagenomics takes DNA and RNA samples found from soil or other samples and reconstructs potential genomes. From genomes, we can predict possible amino acid sequences and from that, we can predict natural proteins that we would have never seen before!!! There is always a risk that since the data is from organisms that are cultivatable that the proteins of others may be different in some way, but that is extremely unlikely to be an issue in my opinion. Very Exciting!! On top of this, if we can predict ligand/substrate binding then we can see natural biological compounds that we haven't seen in the lab before! Natural products have a high efficacy in drug discovery so this is fantastic!
Big things that still need to be done, in the field are stability prediction, function prediction, ligand/substrate binding prediction, product prediction, dynamic movement prediction for enzymes that go through conformational changes. Lots ahead, but breakthroughs like this make me super hopeful!
You're an inspiration Lex! I am hoping to make an impact on substrate/product prediction myself with applications of deep learning!
Thank you for everything you do Lex!
Wow that's exactly the AlphaFold coverage I needed!!
Thank you so much Lex, you are a beast!
Excellent summary insight. Thanks for a guide thru this most complex field.
These short videos are fun for us too. Thank you Lex for teaching us in a way we can easily understand.
Really impressive work by Lex in just a few days. I get a lot out of this and hope to see more regarding this breakthrough. Particularly I realized that I need to learn more about transformers. I'm not sure what it means but seems similar to the idea that has been on my mind for a while regarding the determination of relevant features...
Mind was blown when I first read this. I think a lot of people don't quite understand just how amazing this is (or at least will be in the near future). I'm not in drug design etc for now, but I work in medicine. This is completely world-changing for so many entities out there depending on crystallography or just trial and error.
Researcers from DeepMind have the greatest contribution to this major achievement by finding the right approach for such a complex process to be sorted out
Please keep doing this type of videos! Very informative and fun!!!
Keep doing this kind of videos. They are amazing!
@Lex Fridman Keep trying to scam people, fake Lex.
Great video!! Please keep doing these type of video's. As an AI student at a university in the Netherlands it's great to keep up to date with reality in this way, thank you.
CRISPR-Cas9 was huge, and we all knew it’d win the Nobel in chemistry or medicine at some point. This is just as big, and could win chem, medicine or physics. If your work is mentioned in three Nobel fields, I can’t even find words for how amazing what you’re doing is. This will change our lives forever even if we don’t notice.
Love the quick videos, please make more!
Just one thing about misfolding. Misfolding are of not too big interest to most molecular biologists. It does play a role in some diseases, but most of diseases are caused by other things. Biologists just want to know the structure to know how it interacts with other molecular machinery, be it healthy or unhealthy parts, or in pathogens. The 3D structure is just useful for many things, be it progressing knowledge, drug design, etc. Also a lot of proteins are composed of more than just amino acids. They can have so called post-translation modifications (happens after translation or during), that do include extra attached groups, be it non-amino acids, or short peptide groups, or sometimes non-peptide groups (called cofactors or prostethic groups), thus forming a bit branching strcture (kind of tree in CS), or be composed of multiple sub-units of the same or different type, that together create some weird complexes. All of them determine 3D structure, stability, folding and function. So the complexity is actually way higher than one would thing just by looking at the sequence. I don't belive sequence itself fully describe the protein. It does in many cases, but not always. Anyhow, biologists just want to know the structure of correctly folded protein for many reasons. Study of misfolding is rather niche, but of course important to some stuff, and progressing knolwedge of the folding process itself. Also I think the 10^143 figure, is just deceptive. It is not like nature tries all the combinations. Proteins are mosly synthetized in sequential process, one amino acid at the time, not in parallel, so obviously, the folding process is mostly incremental, and determined or rather influenced by the enzymes that do the synthesis - folding happens during synthesis naturally. It is like saying air molecules in the room I am in can be in one of 10^100000000000000000000000000000000 possible configurations, but it always finds the energy minimum and spreads uniformly. Or that there might be 10^1000000 books possible, yet humans, always write ones that do have a structure and sense and can be parsed. It silly to even mention this paradox, because it is not a paradox.
Make more, these quick videos are Awesome.
Thanks for the video Lex! I couldn't be more excited about this breakthrough. The profound implications for the future (years, not decades) of health really hit me in a deep way and make the AI revolution feel like it's well and truly underway.
I'm doing PhD study in Computational Chemistry and I'm using ML code to research new area of materials. I haven't got results yet but it performs extremely better in most of side than conventional modelling method so far.
WOW! Thanks for bringing this to my attention.
I vote more videos like this. Good job Lex
These short videos are definitely useful Lex
This quick videos are great, please keep making them.
Nice, keep doing more of this short videos 👌
@Lex Fridman Keep trying to scam people, fake Lex.
@@TripleDot-w4s Keep up the good work!
Quite literally the definition of groundbreaking work.
you are the best thing to happen to youtube in years! Love the podcast!
Lex, I truly appreciate your work brother.
Really appreciate you sharing what you learn. Thank you very much!
Very inspiring! Definitely going to spend some time today reading up on protein folding. Thanks for the lesson.
@Lex Fridman You've been hacked lol
This is a great video with a useful insight into how DeepMind came to win CASP. As a biochemist and bioinformatician who has been part of the effort to "solve the protein folding" problem I would say that strictly speaking the problem has not actually been solved. I say this because the research many people focused on was "how do proteins fold into specific configurations?" The DeepMind solution does NOT answer that question. There is no insight into the biophysical rules that guide the folding. And the reason that research has centred specifically on understanding "how" is because the answer to THAT question gives answers in BOTH directions of the problem. DeepMind can go from sequence to structure, but it cannot go from structure to sequence ("I want a protein that has this structure - what sequence gives me that?").
Of course, with further AI and computing power, it is possible to start from a structure that has a similar form to what you wish to create and therefore it is just a matter of applying a genetic algorithm which does what it was created to mimic, to go through the combinations until the target structure is reached. But still, it isn't "understanding", it is just doing.
The other thing I would say that DeepMind does not do is offer insight into diseases caused by misfolded proteins. And to be specific here, a misfolded protein is a protein that has not folded into the configuration that its sequence predicts to fold into, usually because of unusual factors within the folding environment. DeepMind can only go from sequence to structure. It cannot provide any information about the folding process or influences of the folding environment. For diseases where "misfolding" is due to sequence mutations strictly speaking the protein is folding correctly, just not the same as the native sequence. For these mutated sequences, yes, DeepMind will help.
I have said for a long time that the problem of protein structure prediction would be broken by the brute force of AI. But the protein folding problem still exists.
Great vid lex. Short and to the point.
Love this sort of content, Lex, please do more
@Lex Fridman Keep trying to scam people, fake Lex.
Very useful. Please make more!
I would love to get my hands on Alphafold2 for my research someday. Please keep making more videos about this! Thank you Lex.
@Lex Fridman Keep trying to scam people, fake Lex.
I'm feeling old.
I remember when "solving the protein folding problem" was a catchphrase for some far off, indefinable future; and now here we are.
Great overview. Please make more short videos. They help me immensely
@Lex Fridman Keep trying to scam people, fake Lex.
thank you for the black background
this is so fascinating.
machine learning, AI, genetics, epigenetics, quantum computers, fusion...
my greatest wish is that i live long enough to see many of these future developments
Now it makes sense. Thanks Lex.
Thank you, great video. Please keep doing such videos!
@Lex Fridman spam?
@@lpandp90 Dude! I got the same exact message. I've been scrolling through the comments looking for any other responses by him.
@@lpandp90 It's probably spam
One of the better reports on this breakthrough, thanks Lex. In reality there are really only 20 naturally occurring proteinogenic amino acids since Selenocysteine which is homologous to cysteine replaces a Sulfur atom with Selenium and is only found in about 50 proteins. X-ray crystallography will still be around where there is a need to solve more complex protein structures where subdomains exist or to understand how small molecule drugs bind to protein targets. Maybe that will be the next breakthrough but a much tougher problem since there is not a big training set for AI. It will be interesting to know how well this methodology can be applied to predict variations in protein folding brought about by mutations in protein sequences and other important applications such as designer proteins if that is possible.