EvolutionaryScale

New! Announcing new frontier biological foundation models on AWS in a collab with the amazing EvolutionaryScale. The opportunity for AI to improve health care is huge: it will be one of the biggest drivers of health and wellness improvements this decade - accelerating the momentum behind precision care, drug discovery, and biological research and development by several orders of magnitude. Today, we're bringing this reality one step closer through a collaboration with EvolutionaryScale, bringing their new "ESM3" biological AI models to everyone on AWS. Let's take a closer look… 🧬 Our biology is made up of systems - billions of vibrating, chaotic, multi-dimensional, densely-packed, beautiful, complex, interconnected systems - each one nudging countless others to stay in balance with one another. 🩵 AI offers perhaps the largest opportunity in a generation to build new ways of understanding and simulating these chaotic systems, allowing us to precisely manage them and tip the scales in favor of health and wellness. 🌤️ Today, we are announcing a partnership with EvolutionaryScale, and the availability of new cutting edge “ESM3” biological AI on AWS. 💭Just how LLMs learn to understand the relationships between words and topics, and can interpret a prompt to express them in different modalities, so ESM3’s AI models learn to understand relationships between sequence, structure and function, and can interpret a prompt to express them in different modalities. 💬 ESM3 can reason over the sequences, structures, and functions of proteins to generate new protein structures - with atomic level accuracy - which "fit" within the complex, multi-dimensional systems of real biology while meeting the precise functional requirements of the prompt. 🙌 Using models like these, there is an opportunity to speed up drug development by designing active molecules more quickly and efficiently, create more personalized treatments, drive down antibiotic resistance, improve diagnostics and disease detection (especially for rare diseases), create new vaccines, and improve chronic disease management. 🫀 Over time, these models will not only understand how biological molecules function together, but also… …how cells interact with other cells in tissues… …how tissues interact with other tissues in organs… …and how our organs interact with each other in our own bodies… …leading the way for sophisticated new ways to find health and wellness through balance. It'll take a beat, but it's a very exciting, promising new area. ESM3 is available on SageMaker and HealthOmics today, so it can be incorporated into health and life sciences drug discovery workflows immediately. We'll also add ESM3 to Bedrock later this year. Fire it up and prompt healthy. PS: Below, a new fluorescent protein designed by ESM3. It has the same function as the protein which makes some coral glow, but in an entirely new sequence - more novel than it is similar to existing proteins.

Delighted to co-lead the $142M seed round in EvolutionaryScale (now out of stealth) with Nat Friedman, Daniel Gross, and Lux Capital, a new frontier AI lab that has now trained ESM3, a natively multimodal and generative language model for proteins and the largest biological language model to date ESM3 jointly reasons over protein sequence, structure, and function and launches with NVIDIA and Amazon Web Services (AWS) partnerships. Check out the preprint here, which simulates 500 million years of evolution and enables generative programming of proteins: https://lnkd.in/gBH4ThKU The ESM team, previously Meta, was the first to describe scaling laws for biology, following similar observations in LLMs for language and vision. ESM3 reveals an new architectural breakthrough, encoding 3D atomic structure as discrete tokens for controllable protein design Congratulations to Alex Rives, Salvatore Candido, Tom Sercu, and the rest of the 10X engineers and researchers at EvolutionaryScale! Our Arc Institute lab is proud to be a collaborator on the ESM3 study, with contributions from Vincent Tran and Liam Bartie.

NEW: A group of ex-Meta scientists have raised a $142 million seed round to advance AI research, particularly a newly unveiled model that can consider the sequences, structures, and functions of proteins. EvolutionaryScale's seed round was led by Lux Capital and the duo of tech founders-turned-investors: Nat Friedman and Daniel Gross. The startup also released a preprint on Tuesday of ESM3, the latest generation of the protein language model it had initially developed at Meta. The group was able to create a new glow-in-the-dark protein, effectively simulating evolution with the technology. Chief scientist Alex Rives estimated would take 500 million years to happen in nature. The long-term vision is a sweeping one, spanning not just therapeutics but environmental and industrial applications for new proteins, like ones that can break down plastics or capture carbon. “We will really be able to think much more creatively about designing biology and will have the tools to be able to approach biology not through painstaking trial and error, but through engineering from first principles,” Rives said. My latest in Endpoints News with Ryan Cross: