Biology-focused artificial intelligence startup EvolutionaryScale Inc. said today it has raised $142 million in a seed funding round to transform the way scientists create novel proteins for scientific research.

Today’s round was led by Nat Friedman, Daniel Gross and Lux Capital, and saw participation from Amazon Web Services Inc. and Nvidia Corp.’s venture capital arm.

The startup has developed what it claims is the first large language model geared toward the design and creation of new proteins and other biological systems that can accelerate drug discovery, and other use cases, such as engineering microbes that can break down plastic in the environment.

In an interview with Reuters, EvolutionaryScale’s co-founder and Chief Scientist Alex Rives said that he and his colleagues started out developing generative AI models for proteins while working at Meta Platforms Inc.’s AI research team FAIR in 2019. He left Meta after his team was disbanded, and along with Tom Sercu and Sal Candido, founded EvolutionaryScale to continue that work.

By simulating novel proteins, it’s possible to reveal the mechanisms of diseases and identify ways to slow down their progress or even reverse them. Meanwhile, by creating new proteins, scientists can potentially develop entirely new drugs and therapeutics. However, the existing processes for designing proteins in a laboratory are both slow and expensive.

The challenge of designing a protein is that researchers must first create a structure that could potentially perform a specific function in the human body, and then find an amino acid sequence that can actually fold into that structure. Proteins must fold correctly into very specific three-dimensional shapes so they can perform the intended function.

New proteins evolve over millions of years, and EvolutionaryScale aims to speed up this process. Its ESM3 LLMs have been trained on a dataset that contains the structures and sequences of more than 2.78 billion proteins and can perform advanced reasoning to design entirely new ones.

The startup said it’s making the largest, 98 billion-parameter version of ESM3 available for commercial uses via its cloud-hosted Forge platform, which will launch on AWS’ cloud first. It has also released a smaller version of the ESM3 model for researchers who don’t intend to commercialize their work.

EvolutionaryScale plans to make money through a series of licensing fees, partnerships and revenue-sharing agreements. For instance, it hopes to work with pharmaceutical companies to integrate ESM3 into their drug design processes, or share revenue with research teams that made breakthrough discoveries using its models.

In addition to launching on AWS, the startup’s platform will also be made available to select customers through Nvidia’s NIM microservices offering.

The startup revealed that it has already used ESM3 to create a new kind of green fluorescent protein or GFP, which is the family of proteins responsible for luminescent colors in nature, such as glowing jellyfish and coral. It said that this new protein would have taken up to 500 million years to evolve naturally.

Holger Mueller of Constellation Research Inc. said the complexity of protein engineering has held back scientific progress for a long time, as existing computing architectures simply haven’t been up to the task of doing this at anything like the speeds it needs to be done.

“This is changing with the rise of generative AI, with startups like EvolutionaryScale creating a model that can help scientists to understand and model complex proteins far more quickly,” the analyst said. “The ESM3 model has already seen some uptake, with backing from the likes of AWS and Nvidia. It’ll be interesting to see what kinds of scientific breakthroughs result, hopefully in the near future.”

EvolutionaryScale isn’t the only company looking to use AI to accelerate protein discovery. Google LLC’s DeepMind is performing similar work with its AlphaFold LLMs, while OpenAI has partnered with the French pharma company Sanofi SAS to advance drug development with its LLMs.

These companies may need to proceed with caution, though, as there have been warnings from experts that these kinds of biological LLMs could potentially be used to create deadly bioweapons based on novel pathogens and toxins.

The startup said the money from today’s round will go towards training the next generation of its ESM3 model, and growing its partnerships with the biotechnology industry.

Image: SiliconANGLE/Microsoft Designer