EvolutionaryScale, a startup backed by NVIDIA and Amazon, has announced the release of ESM3, a groundbreaking generative AI model for protein design. The model aims to revolutionise biological research and accelerate drug discovery.
The ESM3 model, trained on nearly 2.8 billion protein sequences, represents a significant leap in AI-driven protein engineering. It offers researchers a programmable platform to simultaneously analyse protein sequence, structure, and function.
Tom Sercu, co-founder and VP of engineering at EvolutionaryScale, expressed enthusiasm about the model's capabilities: "In our internal testing we've been impressed by the ability of ESM3 to creatively respond to a variety of complex prompts. It was able to solve an extremely hard protein design problem to create a novel Green Fluorescent Protein."
The 98 billion parameter model utilised NVIDIA H100 Tensor Core GPUs for its development, marking the most substantial computational effort ever invested in a biological foundation model. ESM3 employs approximately 25 times more flops and 60 times more data than its predecessor, ESM2.
EvolutionaryScale's technology has potential applications in various fields, including cancer research, environmental mitigation, and the development of alternatives to harmful plastics. The model's ability to generate new proteins based on user prompts and self-improve its designs could significantly accelerate scientific progress in these areas.
The company is launching an API for closed beta testing, with a small open version of ESM3 available for non-commercial use. The full ESM3 family of models will soon be accessible to select customers as an NVIDIA NIM microservice, optimised in collaboration with NVIDIA.
As ESM3 becomes available on platforms like Amazon Bedrock, Amazon Sagemaker, AWS HealthOMICs, and NVIDIA BioNeMo, it is poised to transform the landscape of protein engineering and drug discovery. The model's potential to accelerate biological research could lead to groundbreaking advancements in various scientific fields.