Key points from article :
Former Meta researchers, led by Alexander Rives, have launched EvolutionaryScale, an AI biotech startup that recently raised $40 million from Lux Capital. The team, previously part of Meta AI’s protein-folding unit, developed an advanced AI model trained on protein molecule data. This model, similar in approach to OpenAI’s GPT-4, predicts protein structures and has generated a database of 700 million potential 3D structures. Such advancements could revolutionize drug discovery, pollution cleanup, and industrial manufacturing. The startup aims to dramatically scale up its AI model to push the boundaries of biology-driven AI.
The field of AI-driven protein folding gained prominence with DeepMind’s AlphaFold, which made a major breakthrough in 2020 by accurately predicting protein structures. EvolutionaryScale’s model, according to a 2022 Science paper, is significantly faster than AlphaFold but slightly less accurate. Despite AI's transformative potential, many pharmaceutical firms remain skeptical that it will lead to a major revolution in drug development anytime soon. The startup, however, is betting that scaling AI models—akin to advancements in natural language processing—will eventually lead to breakthroughs in biological applications.
EvolutionaryScale faces financial and technical challenges. It projects spending $278 million by its third year, with the majority allocated to computing power. Its long-term goal is to develop a general-purpose AI model for biology, capable of designing drugs, engineering "programmable cells" to fight diseases, and creating molecular machines for environmental applications. However, it must first prove its superiority over AlphaFold and other emerging competitors. Rives, serving as interim CEO, may also take up a faculty position at MIT and Harvard’s Broad Institute, potentially shaping the future of AI-driven biological research.
