Valence Discovery Deal Brings Purpose-Built AI/ML to Charles River Labs’ Clients

 Igor Golovniov/SOPA Images/LightRocket via Getty

Igor Golovniov/SOPA Images/LightRocket via Getty Images

Valence Discovery just announced a strategic alliance with Charles River Laboratories to make its artificial intelligence (AI) and machine learning (ML) platform available to Charles River’s clients. The artificial intelligence platform is designed for molecular property prediction, generative chemistry and multiparameter optimization for preclinical drug discovery.

“We’re excited because this allows every Charles River Laboratory client to be AI-enabled without building AI capabilities internally. This democratizes AI, bringing fit-for-purpose, AI-enabled drug design capabilities to companies that otherwise would remain opaque to AI’s potential,” Daniel Cohen, CEO, Valence Discovery, told BioSpace. “Until now, for drug discovery, AI was a tool that was accessed almost exclusively by large organizations or those with in-house AI expertise.”

This partnership is expected to significantly accelerate discovery efforts from hit design through lead optimization and provide greater chemical diversity and faster optimization against complex project-specific design criteria. 

What sets Valence – until recently, known as InVivo AI – apart from other AI technologies is its expertise in low-data settings. Valence pioneered a technique it calls “few-shot learning” in drug design so that accurate predictive models can be obtained despite limited data.

“Traditional deep learning algorithms are data hungry,” Cohen said. “That’s a big problem in typical drug discovery programs where there often is little data.”

Examples include making predictions about novel, poorly characterized targets, and predicting key absorption, distribution, metabolism and excretion (ADME) and toxicology properties during lead optimization.

“Standard deep learning tools historically fail under those constraints,” Cohen said.

In contrast, with Valence’s “few shot” approach, “We’re first teaching the algorithm the underlying rules of chemistry so that it can subsequently extract meaningful information from small amounts of data,” Cohen said.  As an analogy, “Think of a standard deep learning algorithm as a newly-minted Ph.D., and ‘few shot learning’ as a seasoned drug hunter leveraging decades of knowledge to glean insights into novel problems.

“The ‘few-shot learning’ concept is built around getting computers to learn more like humans. People are very good at learning from very few examples, while deep learning tools (generally) are not.”

Valence’s platform also combines active learning with iterative optimization strategies so that only information-laden compounds are selected for synthesis, thus minimizing the number of iterations needed to develop efficacious molecules.

“This technology lets us start with project-specific data sets in the chemical space we’re working in without having to rely on public domain data,” Cohen added.

The AI/ML technology explores featurization schemes for candidate molecules, runs predictive algorithms to explore efficacy, and develops independent models for key pharmacologic criteria. It also has a generative component, which uses that data to inform the design of novel structures predicted to meet the design goals of the discovery program.

Valence’s AI/ML technology concentrates on small molecules. As Cohen said, “Small molecules are the workhorses of drug design. They lend themselves well to deep learning because of the nature of existing datasets, which let us learn structure-activity relationships we can use to design novel compounds.”

There are intractable problems, however, that small molecules can’t typically solve. “AI/ML, though, lets us interrogate novel chemical space more systematically,” and thus perhaps develop solutions. Drug selectivity and how to cross the blood-brain barrier are two ongoing challenges for the industry, for example. “Those are the types of problems that are well-suited to deep learning,” he said.

The Valence team focused on small molecules because, Cohen said, “Our mission is to empower as many drug discovery scientists as possible. We’re getting exciting results,” he said. “Because we spun out of Mila, we’re confident our technology is at the forefront of the field.” (Mila is an artificial intelligence research institute in Quebec, with more than 500 researchers specializing in the deep learning.)

This is Valence’s fourth major partnering announcement since coming out of stealth mode in March.

“We’ve spent the past 18 months doing aggressive R&D, validating and building up the platform. Now our technology is ready for us to begin partnering more broadly to support as many companies as possible with AI-enabled drug design.

“Our partnerships with the Institute for Research in Immunology and Cancer (IRIC), Repare Therapeutics, and Servier are true drug discovery partnerships,” Cohen said. The work with IRIC aims to develop small molecule drugs to treat levodopa-induced dyskinesia in Parkinson’s disease patients, while the Repare project is to develop precision oncology medicines, and the work with Servier aims to develop drugs for multiple targets.

In each case, Cohen said, “We’re providing companies with the ability to accelerate their drug discovery programs, find candidates faster, explore novel chemical space and reduce costs.

“The AI/ML space is still in its early days for drug design,” he continued. “We want to make the technology more accessible across the entire industry.”

The partnership with Charles River Laboratories is another step toward realizing that goal.

Back to news