Latent Labs Inc., an artificial intelligence startup focused on the healthcare sector, launched today with $50 million in initial funding.

Radical Ventures and Sofinnova Partners jointly led the investment. They were joined by more than a half-dozen other backers, including Google LLC Chief Scientist Jeff Dean. Cohere Inc. Chief Executive Officer Aidan Gomez, a former researcher at the search giant who co-invented the Transformer architecture, participated as well.

Latent Labs is led by CEO Simon Kohl, who previously worked at Google parent Alphabet Inc.’s DeepMind machine learning lab. He was part of the team that developed DeepMind’s groundbreaking AlphaFold2 neural network. The algorithm, which helps researchers understand the shape of proteins, earned Google scientists Demis Hassabis and John Jumper one half of the 2024 Nobel Prize for Chemistry.

Proteins, the basic building blocks of life, can twist and fold into various shapes. The configuration of a protein directly influences its behavior. AlphaFold2 significantly reduces the amount of time and effort required to determine proteins’ shape, which enables scientists to speed up medical research.

Latent Labs hopes to take computational biology one step further. The company is developing foundation AI models that don’t merely predict the shape of proteins, but rather generate new protein designs from scratch. It hopes to use those designs to support the development of new medicines.

Latent Labs is focusing its efforts on two types of proteins: antibodies and enzymes. Antibodies help the immune system neutralize harmful particles. Enzymes, meanwhile, are widely used in pharmaceutical  manufacturing because they speed up chemical reactions.

“Every biotechnology or pharmaceutical company wants to be at the forefront of technology to find the best therapeutic molecules, yet not all are in a position to develop the most advanced AI models for the job,” Kohl said. “That’s where Latent Labs comes in.”

A job posting on Latent Labs’ website indicates that it’s building custom software to speed up its AI model development efforts. In particular, the company hopes to automate the so-called hyperparameter search phase of the workflow.

Hyperparameters are settings that influence how an AI model processes data. Those settings define details such as the number of artificial neurons in a model and the way those neurons coordinate their work. Developers typically don’t search for the optimal combination of hyperparameters manually, but rather use algorithms that can quickly try a large number of variations.

Latent Labs is building a wet lab, a type of laboratory optimized to process chemicals, in order to test the proteins developed by its AI models. The lab will also have a second purpose: generating research data that the company can use to fine-tune its AI models.

Kohl told Forbes that the company has already “seen very good success with our own models.” Latent Labs plans to commercialize its models by making them available to researchers for use in drug development. In addition, the company intends to ink “project-based partnerships” with drugmakers that will see it earn commissions when those companies meet key project milestones.

Latent Labs will use its new funding to hire more AI researchers and expand its graphics card cluster.

Photo: Unsplash