A comparatively recent startup called Evolutionary Scale secured a huge tranche of funds to build artificial intelligence models to generate recent proteins for scientific research.
EvolutionaryScale announced today that it has raised $142 million in a seed round led by former GitHub CEO Nat Friedman, Daniel Gross and Lux Capital with participation from Amazon and NVentures, the corporate arm of Nvidia. The company also released ESM3, an artificial intelligence model it describes as a “frontier model” in biology – one that may create proteins for applications equivalent to drug discovery and materials science.
“ESM3 takes a step toward a future in biology where artificial intelligence is a tool for first-principles design, the way we build structures, machines and microchips, and write computer programs,” EvolutionaryScale co-founder and chief scientist Alexander Rives said in a statement.
Rives, along with Tom Secru and Sal Candido, began developing generative AI models to review proteins at Meta’s FAIR AI research lab in 2019. After their team disbanded, Rives, Secru and Candido left Meta to proceed the work that they had began myself.
Characterizing proteins can reveal the mechanisms of disease, including ways to slow or reverse it, while proteins can lead to completely recent classes of medicine, tools and therapeutics. However, the current technique of designing proteins in the laboratory is expensive, each from a computational and human resource perspective.
Protein design involves coming up with a structure that would do a job in an organism or product, and then finding a protein sequence – the sequence of amino acids that make up a protein – that may likely “fold” into that structure. Proteins must fold properly into three-dimensional shapes to perform their intended function.
Trained on a dataset of two.78 billion proteins, ESM3 can “parse” protein sequence, structure and function, Rives says, enabling the model to generate recent proteins along the lines of Google DeepMind’s AlphaFold. EvolutionaryScale makes the full 98 billion-parameter model available for non-commercial use through its Forge cloud development platform and makes a smaller version of the model available for offline use.
EvolutionaryScale says it used ESM3 to generate a recent variant of green fluorescent protein (GFP), the protein responsible for the glow of jellyfish and the luminescent colours of corals. A pre-print on the company’s website provides detailed information about its work.
“We’ve been working on this for a long time and we’re excited to share it with the scientific community and see what they do with it,” Rives continued.
EvolutionaryScale is, in fact, not a charity — the company with about 20 employees tells TechCrunch it plans to make money through a combination of partnerships, user fees and revenue sharing. For example, EvolutionaryScale can partner with pharmaceutical firms to integrate ESM3 into their workflows or share revenues with researchers to develop breakthrough discoveries commercialized using ESM3.
To this end, EvolutionaryScale says it would soon make ESM3 and its derivatives available to pick out AWS customers via the AWS SageMaker AI development platform, the Bedrock AI platform, and the HealthOmics service. ESM3 will even be available to pick out customers using NVIDIA’s NIM microservices, powered by an Nvidia enterprise software license.
EvolutionaryScale says each AWS and Nvidia customers will have the ability to tune ESM3 using their very own data if they need.
It may take some time before EvolutionaryScale becomes profitable. In a company presentation, a copy of which Forbes managed obtained last August, the EvolutionaryScale project repeatedly emphasized that it might be a decade before generative AI models help design therapies. The company will even have to fend off competition equivalent to Isomorphic Labs, a spinoff of DeepMind that already has contracts with major pharmaceutical firms, in addition to Insitro, publicly traded Recursion and Inceptive.
The biggest goal of the EvolutionaryScale project is to scale up model training to include data beyond proteins and create a general-purpose artificial intelligence model for biotechnology applications.
“The incredible pace of new advances in artificial intelligence is being driven by ever-larger models, ever-larger datasets, and growing computational power,” said an EvolutionaryScale spokesperson. “It’s the same with biology. In research over the past five years, the ESM team has explored scaling in biology. We found that as language models scale, they develop understanding of fundamental biological principles and discover biological structure and function.”
That sounds wildly ambitious to this reporter, but having deep-pocketed investors definitely helps.