Recursion and Genentech say they have validated their first drug target discovered through artificial intelligence — a novel neuroscience target found by mapping the biology of lab-grown brain cells at massive scale.

The two have collaborated since 2021 in a partnership worth up to $12 billion in potential milestones. As of August 2026, Genentech has paid Recursion about $216 million, including a recent $3 million milestone tied to this target. Each programme can earn up to $300 million in milestones plus tiered royalties. The specific target and disease were not disclosed.

How the search worked

Recursion built the finding on its Data Factory: a whole-genome CRISPR knockout map made from neurons derived from induced pluripotent stem cells.

The method is systematic rather than hypothesis-led. Every gene in the genome is disabled, one at a time, in separate populations of cells — roughly 20,000 experiments. Each population is then imaged, and the resulting appearance is compared against unmodified cells.

What the images capture is a phenotype: the visible consequence of losing that gene — changes in shape, in the distribution of internal structures, in how cells arrange themselves. Genes whose loss produces similar appearances are inferred to act in related pathways, which is how biological relationships emerge without anyone specifying what to look for.

The scale, and why it is required

More than 1 trillion lab-grown neuronal cells — roughly a dozen brains’ worth — and over 46 million cellular images analysed on the company’s BioHive-2 supercomputer. From initiation to a validated target package took 15 months.

The numbers are not arbitrary. Every gene must be tested in enough independent replicates to distinguish a real effect from experimental variation, and image-based phenotypes are noisy — cells differ in appearance for reasons unrelated to the edit. Statistical power comes from volume.

Why neuroscience specifically

“Finding new targets in neuroscience has historically been challenging,” said Recursion CEO Najat Khan, adding that the milestone “highlights our ability to uncover novel biology in areas where conventional approaches struggled.”

The difficulty is concrete. Brain tissue cannot be biopsied from living patients to study disease mechanisms. Animal models of neurological and psychiatric conditions correspond poorly to the human diseases they represent. And the relevant biology involves cell types that were, until recently, unobtainable in usable quantities.

Neurons derived from induced pluripotent stem cells address that last constraint directly — human neurons, in unlimited quantity, in a dish. Combining them with genome-wide screening is what makes an unbiased search possible at all in this area.

What a validated target is

The term marks a specific and early point in development, and it is worth being precise about what has and has not been established.

A validated target is a protein for which there is evidence that modulating it should affect disease. It does not mean a drug exists, that the protein can be drugged at all, or that acting on it will help patients.

Many validated targets never yield medicines. Some have no binding pocket. Some prove essential to normal function, so inhibiting them causes unacceptable harm. Some behave differently in patients than in cells.

The $3 million milestone reflects that position accurately — a small payment against $300 million available per programme, because most of the risk remains.

The scope of the collaboration

The broader partnership spans 40 programmes and six whole-genome phenomaps, concentrated in neuroscience plus one gastrointestinal-oncology indication.

Forty programmes producing one validated target in five years is a rate worth registering plainly. It may be reasonable given how early most are, and it is considerably slower than the headline partnership value suggests.

How to read the $12 billion

That figure represents everything payable if every programme succeeded through approval and commercial milestones — an outcome no pharmaceutical partnership has approached.

The $216 million actually paid over five years is the meaningful number, and it indicates a substantial but ordinary research collaboration rather than a transformative bet.

The undisclosed target, and why that is standard

Neither the target nor the disease was named, which is conventional in these announcements and worth understanding rather than treating as evasion.

Naming a target before patents are filed and secured hands competitors a free lead. Target identification is the part of discovery that is hardest to protect — the information itself is the asset, and once disclosed anyone can pursue the same protein with their own chemistry.

The consequence for outside assessment is real. Without knowing the target, nobody can judge whether it is genuinely novel, whether others were already working on it, or whether the biological rationale is compelling.

What can be assessed is the partner’s behaviour. A pharmaceutical company paying a milestone has examined the underlying package with access nobody outside has, and its willingness to trigger payment is the only independent signal available — which is a weak form of evidence, and the main one on offer at this stage.

A validated target is an early step, with years of drug development still ahead. It is also a concrete result from a much-hyped promise, which is more than many platform companies have produced. Business news, not investment advice.