Joshua Meier’s Post

New development of AI in this area is going to keep getting more and more exciting!

More exciting news from the world of AI! This week, Chai Discovery unveiled Chai-1, a groundbreaking multi-modal foundation model for predicting molecular structures. Chai-1 excels in key drug discovery tasks, providing unified predictions for proteins, small molecules, DNA, RNA, and covalent modifications. Here’s the TL;DR: Chai-1 delivers state-of-the-art performance across benchmarks like PoseBusters and CASP15 protein structure prediction. Unlike many existing tools, it operates without multiple sequence alignments while maintaining high accuracy. It outperforms previous models in predicting multimer structures, even using just single sequences. Chai-1 can integrate new data, like lab-derived restraints, to enhance performance. The model is freely accessible via a web interface (including for commercial use), with weights and code available for non-commercial use. Overall: Chai Discovery, founded by experts from top AI and drug discovery firms, aims to revolutionize biology by making it an engineering discipline. Getting closer to faster drug discovery and personalized medicine!

OpenAI and Thrive Capital recently backed Chai Discovery, a six-month-old AI biology startup founded by ex-OpenAI and Meta researchers that raised $30 million to develop AI models for drug discovery. 🔬 Chai’s AI model, Chai-1, predicts biochemical molecule structures, potentially speeding up drug development. Chai-1 can work with proteins, small molecules, DNA, and RNA, making it versatile for various applications. 📈 The company claims Chai-1 outperforms Google DeepMind’s AlphaFold on certain benchmarks. 🆓 Chai is making its first model free and open-source for non-commercial use. Google DeepMind’s AlphaFold is the pinnacle of protein prediction models, and Chai is already giving it a run for its money. The more money flowing into drug discovery, the closer we get to curing major diseases using AI — and Chai is even open-sourcing its first model for the benefit of humanity. https://lnkd.in/guvrvyrM

Chai Discovery has launched Chai-1, a multi-modal AI model designed for molecular structure prediction, with significant implications for drug discovery. The model achieves SOTA performance in predicting protein, DNA, RNA, and small molecule structures. In benchmark tests, Chai-1 outperformed several industry models, including a 77% success rate on the PoseBusters benchmark, surpassing AlphaFold3. Unique to Chai-1 is its ability to function without requiring multiple sequence alignments, making it the first model to predict multimer structures using single sequences at AlphaFold-Multimer quality levels. Chai-1 also demonstrates remarkable flexibility, enabling researchers to enhance its predictions with additional lab-derived data, improving accuracy in tasks like antibody-antigen structure prediction. ➜ Check the news at: https://lnkd.in/dVDkf_ZJ ➜ Vendorvise's official Instagram: https://lnkd.in/ebK2KQ2C 🤖 If you enjoy reading my posts about AI and tech news, hit the like button and share with your friends to support my content! #ai #artificialintelligence #ainews #aipowered #news #blog #chai1 #biology #protein #health #aimodel #opensource #google #alphafold #generativeai #aitech #technology #content #socialmedia #future #new #business #techtrends #innovation #launch #success #markets #people #industry

🔹 PertKGE, a new AI tool, decodes compound-protein interactions (CPIs) from transcriptomic data, advancing drug discovery by analyzing gene regulation events. 🔹 Knowledge graph embedding allows PertKGE to simulate complex biological processes, offering insights into both novel drug targets and compounds. 🔹 Key breakthroughs include the identification of ENPP1's role in immunotherapy and novel targets for cancer treatment, showcasing the tool's potential in accelerating pharmacological research. Source: [https://lnkd.in/eC_Wn77p] —————— P.S.: PertKGE’s use of knowledge graphs to model multi-level gene interactions offers a more granular and precise method for predicting CPIs. This approach is likely to push AI-driven drug discovery to new heights, especially in tackling complex biological systems. #TechNews #Productivity #OpenAI #Courses #Learning #Charts #Insights #Analysis #Data #Coding #Enhancement #Tips #DataSecurity #WebDesign #Development #Computer #Programming #Developer #WebDevelopment #Programmer #Robotics #ComputerScience #ArtificialIntelligence #AI #MachineLearning #DataScience #BigData #NeuralNetworks #SoftwareDevelopment #Automation #DataVisualization #Python #DeepLearning #Analytics #DataAnalysis #MachineLearningAlgorithms #AICommunity #MachineLearningEngineer #MachineLearningTraining #AIInnovation #CyberSecurity #NLP #LLM #TextProcessing #NaturalLanguageProcessing #LanguageModel #GPT #Bard #TextAnalytics #SpeechRecognition #Chatbots #PythonCoding #CodingLife #ProgrammingLife #SoftwareEngineer #Code #DataAnalytics #DataMining #Statistics #DigitalTransformation #DataScientist #ComputationalScience #QuantitativeAnalysis #BusinessGrowth #Entrepreneurship #Startup #Innovation #DigitalMarketing #SocialMedia #Marketing #Branding #Economics #Fintech #Security #CloudComputing #Blockchain #IoT #VR #AR #AIForGood #SustainableTech #EdTech #HealthTech

OpenAI and Thrive Capital recently backed Chai Discovery, a six-month-old AI biology startup founded by ex-OpenAI and Meta researchers that raised $30 million to develop AI models for drug discovery. 🔑 Chai’s AI model, Chai-1, predicts biochemical molecule structures, potentially speeding up drug development. 🔑 The company claims Chai-1 outperforms Google DeepMind’s AlphaFold on certain benchmarks. 🔑 Chai-1 can work with proteins, small molecules, DNA, and RNA, making it versatile for various applications. Google DeepMind’s AlphaFold is the pinnacle of protein prediction models, and Chai is already giving it a run for its money. The more money flowing into drug discovery, the closer we get to curing major diseases using AI — and Chai is even open-sourcing its first model for the benefit of humanity. #ai #technology https://lnkd.in/gVjVWgW6

Explore five pioneering startups harnessing AI in drug discovery. From Canadian innovators decoding rare genetic disorders with machine learning to Transatlantic teams optimizing clinical trials, these companies are shaping the future of medicine. Read more: https://lnkd.in/dsvA9TE7 #AI #DrugDiscovery #HealthTech #MachineLearning #DigitalHealth #ClinicalIntelligence #SpeechRecognition #Healthcare

Alphafold 3: new AI diffusion model to predict structure and interactions of all of life’s molecules In a paper published by Google DeepMind, authors introduced AlphaFold 3, a revolutionary AI diffusion model that can predict the structure and interactions of all life’s molecules with unprecedented accuracy. 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗣𝗿𝗼𝗰𝗲𝗱𝘂𝗿𝗲 - Trained on diffusion based architecture, replacing the AlphaFold 2 Evoformer with the simpler Pairformer Module . - New architecture reduces the amount of multiple sequence alignment (MSA) processing  by directly predicting the raw atom coordinates with a Diffusion Module, replacing the AlphaFold 2 Structure Module that operated on amino-acid-specific frames and 71 side chain torsion angles -  distillation training from AlphaFold 2 predictions, and addition of a ranking term to encourage results with more solvent accessible surface area. 𝗔𝗹𝗽𝗵𝗮𝗙𝗼𝗹𝗱 𝗦𝗲𝗿𝘃𝗲𝗿 - released as free platform to harness the power of AlphaFold 3 to model structures composed of proteins, DNA, RNA and a selection of ligands, ions and chemical modifications, in few clicks. 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗮𝗹 𝗔𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 - AI assisted drug discovery and development - map the behaviour for all of life’s molecules, including human DNA 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 𝗥𝗲𝘀𝘂𝗹𝘁𝘀 - exhibits enhanced accuracy in predicting protein-ligand interactions, protein-nucleic acid interactions, and antibody-antigen interactions - outperforms other related models/tools and AlphaFold-Multimer v2.3. 𝗠𝗼𝗱𝗲𝗹 𝗹𝗶𝗺𝗶𝘁𝗮𝘁𝗶𝗼𝗻𝘀  - model outputs do not always respect chirality (property of a figure that is not identical to its mirror image), despite the model receiving reference structures with correct chirality as input features - switch from the non-generative AlphaFold 2 model to the diffusion-based AlphaFold 3 model introduces the challenge of spurious structural order (hallucinations) in disordered regions. - multiple random seeds for either the diffusion head or the overall network do not produce an approximation of the solution ensemble 𝗣𝗮𝗽𝗲𝗿: https://lnkd.in/eaGF9Bq8 𝗕𝗹𝗼𝗴: https://lnkd.in/efawN75s

Building relatable datasets and democratizing AI tools. Yesterday, as part of a panel for Fierce JPM Week, Chief R&D and Chief Commercial Officer Najat Khan, PhD joined Alexandra Gonçalves, MD, MMSc, PhD from Bristol Myers Squibb, Lars Fogh Iversen, PhD from Novo Nordisk, Sreenu Prakash Sai Jasti from Bayer, and Maria Luisa Pineda, PhD. from Envisagenics to discuss “Our AI Future: A Sector-By-Sector Breakdown.” ▪️ Hosted by Zoey Becker from Fierce Pharma, the panel explored the impact of AI in drug discovery and development, and how pharma and TechBio are collaborating and driving innovation in everything from designing novel, highly optimized molecules to building relatable biological datasets. ▪️ Najat pointed to REC-1245 – a highly selective RBM39 inhibitor – as an example of how we leveraged AI to not only identify a key cancer target, but applied precision chemistry design to create a novel drug that inhibits RBM39 without directly impacting CDK12 or CDK13, providing a new mechanism for targeting tumors. ▪️ While we collectively only understand “10-15% of biology,” Najat said, Recursion is actively engaged in building the “readable and connected” datasets to unlock the unknown spaces. We now have 60+ petabytes of proprietary and fit-for-purpose data across phenomics, transcriptomics, ADME, and animal and patient data. ▪️ The panel also provided a great opportunity to share how we’re democratizing access to our AI tools to leverage pharma expertise, including through our partnership with Bayer, the first beta user of LOWE – our “ChatGPT for scientists.” 👉 Learn more about LOWE: https://lnkd.in/ggVHiCHj #JPM25 #JPM2025 #TechBio

The AI hype-cycle has been very focused on LLMs, Image Generators, and recently, AI generated music. These are all novel applications and sure to shake up many industries. But can they save lives? An AI application I'm particularly amazed by is Google DeepMind's AlphaFold, that can predict the structure and interactions of all of life’s molecules. "Given an input list of molecules, AlphaFold 3 generates their joint 3D structure, revealing how they all fit together. It models large biomolecules such as proteins, DNA and RNA, as well as small molecules, also known as ligands — a category encompassing many drugs." Absolutely fascinating.