Boltz PBC, a startup that builds artificial intelligence models for pharmaceutical researchers, has raised $110 million in Series A funding. Bessemer led the round, which Boltz announced Wednesday. The investor group included Greycroft, Andreessen Horowitz, Amplify Partners, F-Prime Capital, Hummingbird and a larger set of new and returning backers.

The company's technology targets small molecules, the building blocks of many drugs, which can only neutralize a pathogen after binding to it. Drug discovery teams typically begin projects by searching for molecules with high binding affinity for a pathogen, reviewing hundreds of thousands of candidates per project in a process that can take months and cost millions of dollars. Boltz's model, BoltzMol-1, generates binding affinity data by reviewing only a few dozen especially promising molecules rather than the hundreds of thousands usually required. Fewer molecules to study means less time spent on analysis.

Binding affinity evaluations do not produce drug discoveries directly. They yield data points called small molecule hits, faint clues that help researchers identify where to focus their efforts. Boltz staffers asked BoltzMol-1 to search for small molecule hits across 10 challenging and therapeutically-relevant targets, which are biological building blocks involved in disease processes, and the company says the model found hits for six of them. Scientists can also adapt BoltzMol-1 to project-specific constraints, such as configuring it to consider only small molecules that can be made from a limited set of available compounds, while evaluating a much larger number of candidates when material procurement is not an obstacle.

Boltz offers a second model, BoltzProt-1, which can generate tens of thousands of protein designs based on data gleaned from small molecule hits and automatically reviews each protein to determine whether it could be used to develop a new drug. AI models usually evaluate protein designs against rubrics that offer only an indirect view of effectiveness, according to the company, while BoltzProt-1 uses more direct metrics. Boltz says the evaluation mechanism showed higher success than several competing models in internal tests.

Boltz sells its algorithms through an application programming interface billed by token usage. The company says customers' token consumption has grown by a factor of 50 since BoltzMol-1 and BoltzProt-1 launched in June, and it counts six of the world's 20 largest pharmaceutical companies as clients.

The new funding will go toward growing headcount and acquiring more computing infrastructure, which the company expects will help it keep pace with leading AI labs such as newly public Anthropic PBC and OpenAI Group. Both have introduced models highly adapted to biology research tasks over the past year. Boltz says it has already begun training a new generation of molecule design algorithms that are showing substantial improvements over its current models.

More company and startup news from TechManNews.