EvolutionaryScale, a startup born from Meta FAIR, has released its ESM3 model, a cutting-edge protein design tool that can revolutionize the field of programmable biology. With funding from Lux Capital, Nat Friedman, and Daniel Gross, as well as investments from NVIDIA and Amazon, EvolutionaryScale is at the forefront of protein engineering. The ESM3 model, trained on over 2.8 billion protein sequences, offers a scalable platform for researchers to discover new proteins and potentially revolutionize drug development, disease eradication, and environmental solutions.
Utilizing NVIDIA H100 Tensor Core GPUs, the ESM3 model is a significant advancement over its predecessor, ESM2, with 98 billion parameters and unprecedented computational power. The model is natively generative, allowing scientists to input structure and function annotations for protein design. This level of generative training across vast amounts of data provides a powerful tool for in silico biological research and protein engineering.
The ESM3 model is set to power the next big advancement in protein biology with NVIDIA BioNeMo, offering protein designers the ability to generate new proteins, improve protein design based on feedback, and design proteins based on specified functionality. EvolutionaryScale is opening an API for closed beta testing, with plans for a non-commercial open version of ESM3 soon to be available. The model will also be integrated into NVIDIA BioNeMo for drug discovery research, providing a transformative tool for protein biologists and designers.
The computational power required to train models like ESM3 is growing exponentially, with the model trained on the Andromeda Cluster using NVIDIA H100 GPUs and Quantum-2 InfiniBand networking. Select partner platforms, including Amazon Bedrock, Amazon Sagemaker, AWS HealthOMICs, and NVIDIA BioNeMo, will have access to the ESM3 model for further research and development.
Overall, EvolutionaryScale’s ESM3 model represents a significant leap forward in protein design and engineering, offering researchers a powerful tool to accelerate their work and explore new possibilities in the life sciences. With its innovative approach to protein discovery and design, the ESM3 model has the potential to drive groundbreaking advancements in drug development, disease treatment, and environmental sustainability.
Article Source
https://blogs.nvidia.com/blog/evolutionaryscale-esm3-generative-ai-nim-bionemo-h100/