A promising startup named EvolutionaryScale has recently secured a significant investment to develop AI models that can generate innovative proteins for scientific research.
On Tuesday, EvolutionaryScale announced it raised $142 million in a seed round, led by former GitHub CEO Nat Friedman, along with Daniel Gross and Lux Capital. This funding also saw participation from Amazon and NVentures, Nvidia’s corporate venture arm. The startup unveiled its latest AI model, ESM3, a "frontier model" for biology that has the capability to create proteins for applications in drug discovery and materials science.
According to Alexander Rives, co-founder and chief scientist of EvolutionaryScale, “ESM3 represents a pivotal step towards a future in biology where AI serves as a tool for engineering at fundamental levels, similar to how we construct buildings, machines, and software.”
Rives, along with team members Tom Sercu and Sal Candido, began their journey into generative AI models to explore proteins back in 2019 at Meta’s AI research lab, FAIR. Following the disbandment of their team, they left Meta to pursue their vision independently.
Protein characterization is crucial for uncovering disease mechanisms, including potential methods for slowing or reversing them. Moreover, generating new proteins can lead to the creation of novel drug classes, tools, and therapeutics. However, the traditional protein design process is expensive and resource-intensive, both computationally and in terms of human labor.
To design a protein, researchers must devise a plausible structure that can perform a specific function in the body and identify a corresponding protein sequence—amino acids that will fold correctly into that structure. Proper folding into three-dimensional shapes is essential for proteins to function as intended.
ESM3 is trained on a dataset comprising 2.78 billion proteins, allowing it to “reason over” protein sequences, structures, and functions, making the generation of new proteins feasible—akin to Google DeepMind's AlphaFold. EvolutionaryScale is offering the full 98-billion-parameter model for non-commercial use on its Forge cloud developer platform and is also releasing a smaller version for offline applications.
The company claims to have successfully utilized ESM3 to generate a new variant of green fluorescent protein (GFP), which is responsible for the glow in jellyfish and vibrant colors in coral. Detailed results of this work are available in a preprint paper on EvolutionaryScale's website.
“We’ve dedicated significant time to this project, and we’re thrilled to share it with the scientific community to see the innovations they will develop,” Rives expressed.
While EvolutionaryScale has ambitious goals, it is not a non-profit entity. The company, featuring a team of around 20 employees, aims to generate revenue through partnerships, usage fees, and revenue-sharing models. For instance, it may collaborate with pharmaceutical companies to integrate ESM3 into their workflows or share profits with researchers who make significant discoveries using ESM3.
Looking ahead, EvolutionaryScale plans to make ESM3 and its derivatives available to select AWS customers via the SageMaker AI development platform, the Bedrock AI platform, and HealthOmics service. Additionally, ESM3 will be accessible to select users via Nvidia’s NIM microservices, supported by an Nvidia enterprise software license. Both AWS and Nvidia customers will have the opportunity to fine-tune ESM3 with their data.
It may take time for EvolutionaryScale to achieve profitability. In its pitch deck, reportedly obtained by Forbes last August, the company indicated that it could require up to a decade for generative AI models to contribute to therapy design. Furthermore, EvolutionaryScale faces competition from DeepMind’s spin-off, Isomorphic Labs, which has secured contracts with prominent pharmaceutical companies, as well as competitors like Insitro and the publicly traded Recursion.
EvolutionaryScale's ambitious vision hinges on expanding its model training beyond proteins to develop a general-purpose AI model for various biotech applications.
“The rapid progression of AI innovations is fueled by larger models, extensive datasets, and enhanced computational power,” noted a spokesperson from EvolutionaryScale. “This trend is mirrored in the biological sphere. Our research over the past five years has shown that as language models increase in scale, they gain a deeper understanding of biology's underlying principles, enabling them to uncover biological structures and functions.”
The company's vision is undeniably ambitious, but with a strong backing from investors, it appears poised for significant advancements.
Ranking
