Error generating summary. Please try again later.
The biggest problems in the world might be solved by tiny molecules unlocked using AI. Take your big idea online today with https://ve42.co/hostinger - code VE at checkout. A huge thank you to John Jumper and Kathryn Tunyasuvunakool at Google Deepmind; and to David Baker and the Institute for Protein Design at the University of Washington for their invaluable expertise and explanations. We’re incredibly grateful to Juanita Bagawan and Lucie Kerley at Google Deepmind, and Ian C. Haydon at the UW Institute for Protein Design for their assistance with some of the animation and imagery used in this video. Special thanks to those at Google DeepMind Production Studios for their help interviewing John Jumper: Director - Bernardo Resende Director of Photography - Robert Messere Senior Audio Technician - Perry Rogantin Senior Production Coordinator - Sarah Ellen Morton Studio Manager - Nicholas Duke For more on this topic head over to our Patreon for the exclusive extended interview https://www.patreon.com/posts/patreon-with-cut-122274530?utm_medium=clipboard_copy&utm_source=copyLink&utm_campaign=postshare_creator&utm_content=join_link ▀▀▀ 0:00 How to determine protein structures 3:50 Why are proteins so complicated? 5:34 The CASP Competition and Deep Mind 9:08 How does Alphafold work? 12:06 3 ways to get better AI 14:24 What is a Transformer in AI? 17:15 The Structure Module 18:35 Alphafold 2 wins the Nobel Prize 20:36 Designing New Proteins - RF Diffusion 22:58 The Future of AI ▀▀▀ Try Snatoms! A molecular modelling kit I invented where the atoms snap together. https://ve42.co/SnatomsV Sign up to the Veritasium newsletter for weekly science updates - https://ve42.co/Newsletter ▀▀▀ References and Credits: How AI Revolutionized Protein Science, but Didn’t End It. (Jun 2024) via Quanta Magazine - https://ve42.co/airev Jumper, J., Evans, R., Pritzel, A. et al. (Jul 2021). Highly accurate protein structure prediction with AlphaFold. Nature - https://ve42.co/alphafoldtwo Senior, A.W., Evans, R., Jumper, J. et al. (Jan 2020). Improved protein structure prediction using potentials from deep learning. Nature - https://ve42.co/alphafold R Zwanzig, A Szabo, B Bagchi (Jan 1992). Levinthal's paradox. PMC, National Institute of Health - https://ve42.co/levparadox Khatib, F., DiMaio, F., Foldit Contenders Group. et al. (Sept 2011). Crystal structure of a monomeric retroviral protease solved by protein folding game players. Nature - https://ve42.co/proteinfoldingsolved DeepMind co-founder: Gaming inspired AI breakthrough. (Dec 2020) via BBC - https://ve42.co/aibreakthrough 3Blue1Brown. (Apr 2024). Transformers (how LLMs work) explained visually | DL5 via Youtube - https://ve42.co/transformers AlphaFold impact stories. via Deepmind Blog - https://ve42.co/foldimpact Quanta Magazine. (Oct 2024). How AI Cracked the Protein Folding Code and Won a Nobel Prize via Youtube - https://ve42.co/proteinfolding Looking Glass Universe. (Apr 2024). how AlphaFold *actually* works via Youtube - https://ve42.co/foldworks Google DeepMind. (Nov 2020). AlphaFold: The making of a scientific breakthrough via Youtube - https://ve42.co/alphafoldvideo Nobel Prize. (Dec 2024). 2024 Nobel Prize lectures in chemistry | David Baker, Demis Hassabis and John Jumper via Youtube - https://ve42.co/nobelchem The AlphaFold2 Method Paper: A Fount of Good Ideas. (Jul 2021) via moalquraishi.wordpress - https://ve42.co/Alphafoldpaper Why Don't We Have More Protein-Protein Drug Molecules? (Dec 2009) via Science - https://ve42.co/proteindrugmol Data sonification. via Carla Scaletti - https://ve42.co/datasonification Images & Video: Available here - https://ve42.co/alphafoldvisuals And if you want to learn more about how we have been making antivenom check out our previous video How Horses Save Humans From Snakebites - https://ve42.co/veantivenom ▀▀▀ Special thanks to our Patreon supporters: Adam Foreman, Albert Wenger, Alexander Tamas, Anton Ragin, Autodidactic Studios, Bertrand Serlet, Blake Byers, Bruce, Dave Kircher, David Johnston, Evgeny Skvortsov, Garrett Mueller, Gnare, gpoly, Greg Scopel, I. H., Juan Benet, KeyWestr, Kyi, Lee Redden, Marinus Kuivenhoven, Matthias Wrobel, Meekay, meg noah, Michael Krugman, Orlando Bassotto, Paul Peijzel, Richard Sundvall, Sam Lutfi, Spilmann Reed, Tj Steyn, TTST, Ubiquity Ventures, wolfee ▀▀▀ Directed by Sumeet Kulkarni Written by Sumeet Kulkarni, Derek Muller and Zoe Heron Edited by Jack Saxon Animated by Ivy Tello, Emma Wright, Andrew Neet and Fabio Albertelli Illustrated by Jakub Misiek Filmed and demos done by Sumeet Kulkarni Additional research by Geeta Thakur, Vincent Cheng and Emilia Gyles Produced by Sumeet Kulkarni, Derek Muller, Zoe Heron, Rob Beasley Spence and Tori Brittain Thumbnail contributions by Ren Hurley, Peter Sheppard and Ben Powell Additional video/photos supplied by Getty Images and Storyblocks Music from Epidemic Sound
This is the inside story of how David Baker, Demis Hassabis and John Jumper won the 2024 Nobel Prize in Chemistry for advances in computer-assisted protein design and structure prediction. Proteins are biological nano-machines that perform a vast array of vital functions inside of every living organism. For more than half a century, scientists have looked to solve the central mystery of protein science: How does a one-dimensional string of molecules fold innately and near-instantaneously into a complex three-dimensional shape? In 2020, Google DeepMind entered a deep-learning algorithm called AlphaFold2 into the Olympics of protein folding — and to everyone’s shock and surprise, ended up solving a key part of the puzzle. This breakthrough kick-started an AI revolution in biology research, clearing the path to revolutionary new techniques in protein design, which is the process of creating new and novel proteins that could solve some of the world's biggest problems. Read the related article at: https://www.quantamagazine.org/how-ai-revolutionized-protein-science-but-didnt-end-it-20240626/ Related Papers: - "Highly accurate protein structure prediction with AlphaFold" https://www.nature.com/articles/s41586-021-03819-2 - "De novo design of protein structure and function with RFdiffusion" https://www.nature.com/articles/s41586-023-06415-8 - "Generalized biomolecular modeling and design with RoseTTAFold All-Atom" https://www.science.org/doi/10.1126/science.adl2528 - "Accurate structure prediction of biomolecular interactions with AlphaFold 3" https://www.nature.com/articles/s41586-024-07487-w CORRECTION: The protein shown at 01:07 labeled SEROTONIN is mislabeled, it is in fact "Crystal structure of serotonin 2A receptor in complex with serotonin" https://www.rcsb.org/structure/7WC4 --------- Chapters: 00:00 - Introduction 01:03 - What is a protein? 02:31 - Levinthal Paradox 02:53 - The Protein Folding Problem - how proteins fold to function 03:48 - John Kendrew / using X-ray crystallography to determine structure 05:02 - The Protein Data Bank (PDB) 05:45 - Christian Anfinsen's Nobel winning research 06:28 - Chemical structure of amino acids 07:17 - Secondary and tertiary folding structures 07:59 - Quaternary folding structure 08:16 - The beginnings of computational biology 09:09 - Critical Assessment of protein Structure Prediction (CASP) challenge 10:26 - Baker lab develops RoseTTA 11:31 - Google DeepMind introduces deep learning with AlphaGo 12:00 - DeepMind develops AlphaFold 1 to enter CASP 13 13:32 - AlphaFold 2 explained 15:28 - DeepMind wins CASP 14 and solves the protein folding problem 17:10 - An AI revolution in biological research 17:45 - How the Baker lab designs new proteins 19:53 - New AI tools predict cellular interactions, AlphaFold 3 and RoseTTAFold All-Atom 21:23 - David Baker, John Jumper, and Demis Hassabis win the Nobel Prize --------- VISIT our website: https://www.quantamagazine.org LIKE us on Facebook: / quantanews FOLLOW us Twitter: / quantamagazine Quanta Magazine is an editorially independent publication supported by the Simons Foundation: https://www.simonsfoundation.org
Can AI help us model biology down to the molecular level? Neil deGrasse Tyson, Chuck Nice, and Gary O’Reilly learn about Nobel-prize-winning Alphafold, the protein folding problem, and how solving it could end disease with AI researcher and Chief AI Officer of Isomorphic Labs, Max Jaderberg. We break down the protein folding problem and why it was so hard to crack. How do we learn to decode how strings of amino acids fold into complex, dynamic shapes that power our biology? How did AlphaFold crack this decades-old puzzle? We explore how Alphafold 3 uses the shapes of proteins to predict protein-drug interactions. How close are we to using AI and deep learning to design new medicines instead of discovering them by chance? Could generative AI invent molecules that precisely target the cause of a disease? We take these ideas into the future: Could AI help us fix misfolded proteins that cause illness? Can we use these tools to create personalized drugs tailored to your unique genome? Will it make it economically feasible to treat rare diseases? Could this even help us slow aging or regenerate limbs, like salamanders do? Could this technology combat cancer, fix rare genetic disorders, or even digest plastic? And with quantum computing on the horizon, what happens when we combine it with AI? Will chemistry ever be done the same way again? Thanks to our Patrons Riley r, pesketti, Lindsay Vanlerberg, Andreas, Silvia Valentine, Brazen Rigsby, Marc, Lyda Swanston, Kevin Henry, Roberto Reyes, Cadexn, Cassandra Shanklin, Stan Adamson, Will Slade, Zach VanderGraaff, Tom Spalango, Laticia Edmonds, jason scott, Jigar Gada, Robert Jensen, Matt D., TOL, Thomas McDaniel, Sr., Ryan Ramsey, truthmind, Aaron TInker, George Assaf, Dante Ruzinok, Jonathan Ford, Just Ernst, David Eli Janes, Tamil, Sarah, Earnest Lee, Craig Hanson, Rob, Be Love, Brandon Wilson, TJ Kellysawyer, Bodhi Animations, Dave P., Christina Williams, Ivaylo Vartigorov, Roy Mitsuoka (@surflightroy), John Brendel, Moises Zorrilla, deborah shaw, Jim Muoio, Tahj Ward, Phil, Alex, Brian D. Smith, Nate Barmore, John J Lopez, Raphael Velazquez Cruz, Catboi Air, Jelly Mint, Audie Cruz for supporting us this week. Timestamps: 00:00 - Introduction: Max Jaderberg 3:22 - Deeplearning & Neural Networks 8:36 - The Protein Folding Problem 11:45 - Alphafold & Modelling Protein Structure 16:42 - Using AI for Drug Discovery 23:04 -The Root of All Disease 27:00 - Upending the Pharmaceutical Industry 29:35 - Bespoke Medicine 33:45 - Upending Chemistry 35:21 - Can We Model an Entire Human? 39:05 - Upgrading for Space 43:38 - Less Side Effects 44:36 - Modelling with Quantum Computing & More 46:50 - Guardrails & Regulation Check out our second channel, @StarTalkPlus Get the NEW StarTalk book, 'To Infinity and Beyond: A Journey of Cosmic Discovery' on Amazon: https://amzn.to/3PL0NFn Support us on Patreon: https://www.patreon.com/startalkradio FOLLOW or SUBSCRIBE to StarTalk: Twitter: http://twitter.com/startalkradio Facebook: https://www.facebook.com/StarTalk Instagram: https://www.instagram.com/startalk About StarTalk: Science meets pop culture on StarTalk! Astrophysicist & Hayden Planetarium director Neil deGrasse Tyson, his comic co-hosts, guest celebrities & scientists discuss astronomy, physics, and everything else about life in the universe. Keep Looking Up! #StarTalk #NeildeGrasseTyson
This video from the New England Journal of Medicine is a companion to the article "A Holy Grail — The Prediction of Protein Structure": https://nej.md/3PpWv4F
AI-driven protein design, synthetic biology, and computational biology are transforming molecular biology by enabling rapid protein evolution using advanced models like esm3. These innovations in bioinformatics and biotechnology are paving the way for breakthroughs in personalized medicine and sustainable science solutions. #AIProteinEvolution #Biotechnology #SyntheticBiology #SustainableScience #FutureOfMedicine REFERENCE: Simulating 500 million years of evolution with a language model. Science (2025). https://doi.org/10.1126/science.ads0018 Key Points: 🧬 Evolutionary Acceleration: AI models like esm3 compress millions of years of protein evolution into mere days, allowing scientists to explore a vast array of protein possibilities beyond nature’s limitations. ⚙️ AI-Powered Protein Design: The model generates novel proteins by analyzing billions of sequences and predicting structures with enhanced functionality, offering unprecedented potential in drug development and material science. 🧪 Fluorescent Protein Innovation: esm3 successfully created a fluorescent protein only 58% similar to natural ones, demonstrating its ability to generate entirely new protein structures with unique properties. 🌱 Environmental Applications: AI-designed proteins can break down plastics, improve crop resilience, and contribute to sustainable solutions for pressing ecological challenges. 🏥 Revolutionizing Medicine: Tailored proteins designed by AI can target diseases with high precision, accelerating drug discovery and advancing personalized healthcare solutions. ⚖️ Ethical and Regulatory Concerns: The rapid advancement of AI-driven biology raises ethical questions regarding accessibility, security, and potential misuse, necessitating responsible regulation and oversight. 🚀 Future Prospects: With further AI and computing advancements, such as quantum computing, the potential for smart, adaptive proteins capable of responding to environmental changes is on the horizon. Context timestamps: 00:00 - Introduction to AI-driven Protein Evolution 02:30 - Understanding Protein Structures and Functions 05:10 - Novel Proteins and Their Applications 07:45 - Ethical Implications and Future Outlook AI-driven Protein Evolution: Unlocking the Potential of Synthetic Biology!
Google's AlphaFold won a Nobel Prize for revolutionizing biology...so why is the successor locked behind closed doors? An engineer's breakdown of the science, biosecurity concerns, and why our most altruistic AI isn't publicly available. My controversial take: you can't hate AI until you understand it, so let's change that by doing a deep dive into a little corner of it. Today we take a deep-dive look at the AI that just replaced the Nobel Prized AlphaFold model and learn why the company behind its successor refuses to let anyone touch it (wtaf?). We'll also tackle unphysical hallucinations, how a flip-flopped molecule can turn a cure into a poison, and NeuralPLexer3: the model that's "too good" to release. NOTE: All of my videos, despite being on the topic of AI and Machine Learning, are 100% human made, including scripting, editing, and audio. We are interested in positive applications of Machine learning on this channel, and imo that doesn't include creative uses for AI. Humans are for creating. * (This doesn't include any AI pre-generated images that I use as examples of AI generated images, and they are not images I've ever generated myself.) Video Timestamps: 0:00 Intro 1:50 How Proteins Work 3:54 Levinthal's Paradox 4:22 CASP 5:25 Is AlphaFold Dead? 6:02 NP3 (Neural PLexer3) 7:27 Unphysical Hallucinations 10:20 AI Uncanny Valley 11:40 NP3 Decoded 14:46 How Do We Know It Actually Works? 15:42 The Results 16:17 The Trade-Off 17:35 Apo and Holo States 18:22 Cryptic Sites 20:15 Kinases 21:14 What's the Cost? 26:08 The Biosecurity Lockdown 27:16 Limitations 27:47 My Question for You primary sources: https://docs.google.com/document/d/11pafLrYT7ZXJ_7eVvy_GreYqJTS6Yp8qUcGaIU8rsso/edit?usp=sharing 🕯️ Subscribe to come with me. 🥹 My app waitlist: https://www.vyhr.app 📜 Substack (Written Essays): https://substack.com/@rhytomlin 📸 The Visuals: https://www.instagram.com/rhytomlin/ 🪶 CONTEXT I started a company a 8 months ago. I thought being a Founder meant drinking coffee and picking aesthetic fonts. HAHA, what a joke. This definitely isn't a "How To" guide. It's a sort of Lab Notebook, the intellectually messy documentation of building a tech startup from scratch, while aiming to consistently prove that direction beats intensity. Rhy Tomlin v1.0 (startup mode) ABOUT I’m Rhy. The Tech Eldest Daughter (graduate degree in CS, ex-EV industry) who reads Romantasy in between debugging sessions. My mission is to help democratize success by reverse-engineering it (but at the beginning 🙃). I refuse to outsource my curiosity, and I document the messiest middle of building a startup because I believe (some) struggle is the only way to make it useful (and worthwhile, duh). I’m not here to hand out "3 tips to get 6 sales.", okay? I’m here to hand you the schematics as I'm sketching them out so that we can pull the ladder down together. FAQ: What's the app? A personalized mindset tool for people who want to engineer their way through self-limiting beliefs. Current Read: A Tongue so Sweet and Deadly by Sophia St. Germain 📚 Language: Swift Backend: Firebase Design: Figma CMS: Sanity Editor: VS Code (Dracula Theme, obviously) XCode (when I have to 😭)
For over half a century, one problem haunted biology: how do proteins fold? A protein is a chain of amino acids, but its function depends entirely on its 3D shape—like origami made of molecules. Misfolded proteins cause Alzheimer’s, Parkinson’s, and cystic fibrosis. Scientists spent decades using X-ray crystallography and cryo-EM to map structures—one protein at a time, costing millions. Then in 2020, Google’s DeepMind dropped AlphaFold—an AI that predicts protein structures from sequence alone with near-perfect accuracy. Trained on 170,000 known structures, it uses attention-based neural networks to simulate physical and biological constraints. In 2021, AlphaFold 2 stunned the world by solving 98.5% of human proteins—and released a database of over 200 million structures for free. Labs worldwide now use it to design new enzymes, understand disease mutations, and create biodegradable plastics. In 2024, AlphaFold 3 expanded to model protein-DNA, protein-RNA, and protein-ligand interactions—accelerating drug discovery by years. This isn’t just a win for AI—it’s a revolution for medicine. Imagine designing a molecule to disable a virus before it spreads, or engineering bacteria to eat plastic waste. All of it starts with knowing shape. AlphaFold didn’t just solve a puzzle—it handed humanity a master key to life’s machinery. And the best part? It’s open to every researcher, everywhere. 🔔 Hit subscribe for more breakthroughs where AI and science collide to change the world! #STEM #AI #AlphaFold #Biotechnology #scisurge
What if artificial intelligence could help scientists design proteins from scratch—molecules that could one day perform specific functions in biology? 🧬🤖 In this QuarkLore video, we explore the fascinating world of AI protein design, where machine learning and computational biology are being used to generate and study new protein structures. But creating a promising digital design is only the beginning—scientists still need to understand whether these proteins can fold correctly, function as intended, and work reliably in real biological systems. From AI-generated proteins and protein folding to synthetic biology, biotechnology, and computational science, discover how artificial intelligence is opening new possibilities in molecular biology. Could AI-designed proteins transform medicine, biotechnology, and scientific research? Explore the science, possibilities, and challenges behind this rapidly developing field. This AI-created video was manually edited for QuarkLore. Subscribe to QuarkLore for cinematic stories covering AI, science, biotechnology, space, technology, and the mysteries shaping our future.
Google DeepMind has developed a new AI system called AlphaProteo that designs novel protein binders with high binding affinities. These binders have potential applications in various fields, including drug discovery, disease understanding, and biosensor development. AlphaProteo utilizes a large dataset of protein structures from the Protein Data Bank (PDB) and AlphaFold to predict how proteins interact with each other. The system has been tested on several target proteins, demonstrating its ability to create binders with significant binding strength and success rates. AlphaProteo's success highlights the potential of AI to revolutionize protein design, opening up new avenues for biological and health research.