Genentech advanced the collaboration's first neuroscience target into a joint program, providing early evidence that Recursion's platform can generate novel, biologically validated targets for drug discovery.
A joint team from Recursion and Genentech has identified and validated its first target from an AI-driven, neuroscience-specific map of biology. This target is now being advanced into small molecule hit generation and validation jointly with Genentech as a new early discovery program. This target represents the first of multiple new programs that could emerge from this and other maps that we have built as part of our strategic collaboration.
"The advancement of the first unexplored neuroscience target from our collaboration with Genentech into an early discovery program is an important milestone in a field that has been historically challenging to crack.” said Najat Khan, Ph.D., Chief Executive Officer of Recursion.
Despite the fact that neurological conditions affect more than 3 billion people worldwide, making them the leading cause of illness and disability, new treatments have been slow to emerge. FDA approval rates for neuroscience drugs are less than half those for other therapeutic areas — and drugs targeting diseases of the central nervous system have some of the highest failure rates in medicine, both preclinically and clinically.
Our approach shows that three things long considered difficult to combine in neuroscience — disease-relevant data generation at large scale, foundation models built to turn that scale into disease-relevant insight, and deep, complementary collaboration between a technology platform and a world-class pharma partner contributing biological and computational expertise together with hands-on scientific and data-generation, can now come together in practice. In an area of immense, high unmet need where progress has stalled for decades, proving that combination can work is an industry-defining moment.
To forge a path into novel biology, Recursion, Roche and Genentech initiated a partnership to bring Recursion’s industrialized scale, disease-relevant map-building and map-mining machinery to bear in neuroscience, alongside Roche and Genentech’s world-leading neuroscience expertise, AI/ML and genomics capabilities, and drug development experience.
Our ambition wasn’t to create and test serial hypotheses about how we think neurobiology might work. Instead, we set out to build a biologically-relevant, scaled system for surfacing and evaluating dozens of potential hypotheses in parallel.
Human neurons are among the most technically challenging cell types in drug discovery. They take weeks to mature, requiring carefully controlled conditions, and have historically been too difficult to manufacture reproducibly at the scale needed for genome-wide discovery. To overcome this, Recursion worked alongside Genentech to develop cell manufacturing technologies capable of deriving neurons at scale, ultimately producing over 1 trillion hiPSC-derived neuronal cells.
With this massive bank of neurons, the teams knocked out every one of the genome’s 17,000+ genes, captured more than 30 million neuronal cellular images, and crafted a bespoke foundational model to turn this raw data into an interpretable map of potential targets against a range of neurological diseases.
“We made a strategic plan – where can we really differentiate? Going after new technologies, new drug discovery platforms, big data science, computational approaches – and we found in Recursion an excellent partner,” says Casper Hoogenraad, Vice President and Head of Neuroscience at Genentech.
The jointly developed neuron phenomap allowed foundation models to systematically compare every gene against known drivers of neurological disease. Rather than relying on predefined targets, the models searched the entire genome simultaneously, ranking genes by their biological relationship to disease related perturbations and surfacing novel opportunities supported by high-dimensional evidence. Exploring these relationships one at a time would take years of sequential experimentation and still fail to reveal many of the complex biological connections that only become visible when the entire genome is analyzed together.
The team’s first validated novel target in neuroscience reflects that approach: it’s grounded in human genetic evidence through its association with a well-established driver of neurological disease, an evidence-based starting point for exploring entirely new biology.
"Historically, pursuing novel biology meant accepting a tremendous amount of uncertainty," says Khan. "Our goal was to change that. By starting with biology we already understand and systematically searching for related biology across the entire genome, we can pursue unexplored targets with much greater confidence before we go into validation."
A prediction from a model isn't a candidate target yet: it must first be experimentally tested to establish confidence that the predicted biology is real and relevant to disease. The teams across Recursion and Genentech worked together to build a body of evidence through increasingly rigorous stages of validation, with each set of experiments asking a different biological question.
First came pathway validation: Did the target behave the way the models predicted, strengthening confidence that the computational signal reflected real biology?
Next came functional validation: Did manipulating the target produce meaningful changes in disease-relevant human neurons, including neuronal health, metabolism, and electrophysiology?
Finally came disease validation, the highest bar: Did altering the target change the disease phenotype?
Only targets that consistently demonstrated success across all three stages were advanced.
Working together through the collaboration, the teams reviewed the complete body of evidence generated through a validation research plan. Following successful pathway, functional, and disease validation, Genentech exercised the Validated Target Option, advancing the collaboration's first neuroscience target into a joint Small Molecule Program.
For this program, the next step is using Recursion's AI-native chemistry platform to design and optimize novel molecules against the validated target, transforming a biological discovery into a potential therapeutic candidate.
But this is just the beginning. The value of the Maps of Biology extends far beyond this program. Each map is a rich and reusable biological asset that can continue generating and refining novel therapeutic hypotheses with the goal of discovering, validating, and advancing new targets without rebuilding the underlying dataset. And neuroscience is not the only proving ground: the same approach is already advancing programs with partners across other therapeutic areas, including immunology, inflammation, and oncology — giving Recursion's broader platform multiple, independent shots on goal from a single, reusable engine.
“By combining human disease biology, genome-scale exploration, and rigorous experimental validation with our partner Genentech, we've shown it's possible to validate an unexplored neuroscience target," said Khan. "We believe that's how the next generation of medicines will be built."
