DeepMind AlphaProteo AI: A Gift To Humanity! 🧬
❤️ Check out Lambda here and sign up for their GPU Cloud:
📝 The paper "De novo design of high-affinity protein binders with AlphaProteo" is available here:
Blog post:
📝 My paper on simulations that look almost like reality is available for free here:
Or this is the orig. Nature Physics link with clickable citations:
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Gaston Ingaramo, Gordon Child, John Le, Kyle Davis, Loyal Alchemist, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here:
My research:
X/Twitter:
Thumbnail design: Felícia Zsolnai-Fehér –
#alphaproteo
If this leads to a cure for cancer I’m all for it
As long as you are ok with it 😅
Cancer is 200 problems in a trenchcoat.
Ever seen the “hitler cures cancer” video on YouTube?
There isn’t gonna be cure mate, you know how much money they would loose if there was a cure??
Thatll mean we will miss out on all of the money for cancer, thats a huge amount, if a cure would exist we would miss out on all that money, its better for society for there not to be a cure
I guess I’m smart enough to know I’m stupid
That’s the first step towards progress.
Prion diseases are among the scariest things out there. I hope work like this goes some way to finding treatments if not making conditions like CJD, Parkinson’s or FFI a thing of the past.
Prion disease is misfolded proteins, these people are playing with fire and profit will make them bring to market something that will make them feel like gods and we will learn the harsh way that they are not.
That’s exactly what I want: a protein that binds to misfolded molecules that can repair their shape or something like that. That would be a cure for my mother’s Parkinson.
This is basically the same problem that occurs in Alzheimer’s and Parkinson’s diseases too.
I assume FFI doesn’t stand for foreign function interface. What does it stand for?
@@lionelt.9124 Fatal Familial Insomnia.
This is incredible! This will save countless lives
Drug discovery community: “Hold my beer”
The obvious favourites:
– Drug development
– Plastic recycling
But I want to see cheap enzymes for common industrial processes, so we can upcycle plant waste or change the properties of construction timber without nasty chemicals.
And bioweaponry of course. As long as we keep living by trade and competition instead of sharing and collaboration, we’re doomed
In the meantime, pharmaceutical companies will find ways to delay progress and then profit from it
@@ziad_jkhan average collectivization / famine / secret police enjoyer detected
@@vlazurenko What I’m advocating is a non-authoritarian and fully transparent approach where decisions are made together, based on scientific expertise, so as to maximize access within Earth’s carrying capacity, ensuring that everyone’s needs are met. Resources are allocated based on desirability and availability in the most fair and sensible way possible via dedicated open-source algorithms. Again, no police is required, only comprehensive rehabilitation.
@@vlazurenko Crime, dominance and abhorrent are nothing but symptoms of unmet needs. and existing criminals can actually be resensitized. We need no prisons, judges and laws. Hunter-gatherers had none of those after all, not even money and they thrive for 300k years peacefully by sharing everything, until agriculture created dense populations where scarcity prevailed and forced people to compete. That said, we’re naturally wired to collaborate, not compete.
I am not in biology, and I think I don’t need to be in biology to appreciate this kind of work.
Me too. Not even close to being in that field. But. We all are made of biology. Advances in it are amazing, and this may be huge? Can’t say i understand but it sounds like it.
thanks to deepmind for changing the way biotech is done
Keep in mind that everything must always be experimentally verified and tested. However, the more tools we have, the better. Perhaps one day designing a custom protein will be as simple as designing a DNA primer for PCR. We really need protein dynamics with covalent modifcations as well. Much of the interesting stuff happens via regulatory mechanisms like activing some protein binding site or blocking it, for example.
It would be nice to predict the effects of protein modifications, especially if we get to the point where we can run these affinity calculations against an entire proteom at once. Biology is moving towards Star Trek, one step at a time. I can’t wait to design entire regulatory systems and have a company just ship the required components to you!
It’s crazy seeing that the majority of time I spent on my multi-year Bioinformatics PhD ca. 10 years ago to get to good protein structure predictions is today basically available at the press of a button. That’s so wild and seems almost unreal despite me working with AI professionally every day, but that does isolate me from this feeling at all
Don’t worry – it wasn’t meaningless – I am much closer now to those two fields than you probably are and its my opinion that each every single paper that was published in the past will only help push this idea further ahead (and I mean it literally as well) – All this progress is indeed literally standing on top of the shoulders of past researchers, after all evolution (of anything) at its core is very minute incremental updates.
@@Winter_playsReally? You think you’re closer than the literal 10 year phd? Get over yourself man.
I don’t know what “For seven of the targets, between 9% and 88% of the designs tested in the wet lab were experimentally verified as successful binders” means and I can’t find the source. 9 – 88% seems quite a range, and this begs the question what happened with the designs for the other targets.
2:36
I take this to mean that the ratio of successful designs varies depending on the target, which seems reasonable, some targets are harder than others.
they had seven targets, and made multiple binders for all seven. One of the targets had a 88% success rate (which was the highest of the seven). The lowest success rate was 9%, and the last 5 targets had a success in the range of 10%-40%. But yeah, 9% to 88% is quite a range, but that’s very usual for synthetic biology research.
DeepMind is what OpenAi should be
This is a significant development that should be widely reported in the mainstream media, but it is not. This means that humanity is one step closer to eliminating diseases for good.
However, pharmaceutical companies will find ways to delay progress and then profit from it.
Why cure diseases when you can profit from them?
> protein folding is now a solved problem
Never thought I’d hear those words in my lifetime.
I’m only worried about accidental targeting where the designed protein also binds too tightly to other things and have toxic side effects, such as inventing new and exiting prions that cannot be destroyed. This is a great advancement, but as with all bioengineering, a lot of caution is needed
Two Minute Papers is a such a reliable source of great news! And it looks like that good news is about to go exponential
There’s only one recently I think didn’t do its research literally at all involving VR (but that’s because that’s my field perhaps). I don’t know how accurate it is on other topics, but it does have good presentation.
I would love to see cancer cured in my lifetime having suffered from it myself. What a world that would be
Just means more people will die of heart disease or Alzheimer’s. Don’t get me wrong, it would be nice to extend people’s lifespans by another decade, but I don’t really see it fundamentally changing much.
@@michaelleue7594 If it could cure cancer it could cure those two as well.
So now the race is on to spam the AI, find as many binders as possible and then patent them for the good of humanity, instead of for private Corp.
What a time to be alive!
i remember i heard that the protein structures can be seen by very complicated methods but it cost soo much money just to get one protein structure and then see if it can help to certain cases that be able to have a very good approximation using computers is so fucking awesome as it can lead to analyze structures and then see if can be used in certain scenarios without trial and error that cost millions and millions of dollars without even know if will help.