Whole-Genome Maps, Not Wish Lists: Christopher Winrow on Recursion's First Genentech-Validated Neuro Target

Neuroscience drug discovery has a grim scoreboard. Recursion cites industry data showing that only one in forty neurology candidates that enter Phase I ever reaches approval, and that just 8.4% of programs survive the full clinical gauntlet. Against that backdrop, the company and its partners at Roche and Genentech announced a milestone this week: the first validated neuroscience target discovered through an AI-built phenomics map, now advancing into a co-development program.

This article reconstructs the milestone from public statements by Christopher Winrow, PhD, Recursion's vice president of neuroscience, in an interview published by Genetic Engineering & Biotechnology News on August 6, 2026. Classy AI News did not conduct an independent interview.

The milestone in plain terms

Genentech exercised the collaboration's first "validated target option" after accepting Recursion's validation package. That triggers a $3 million milestone payment and opens a small-molecule early discovery program. Recursion says the payment brings total cash received from Roche and Genentech under the partnership to $216 million since 2021.

The target itself remains undisclosed. What Recursion and Genentech are willing to say publicly is structural: the hit emerged from a whole-genome phenomics map built on human induced pluripotent stem cell–derived neurons, not from a pre-selected pathway list.

"Finding new targets in neuroscience has historically been challenging, and this milestone highlights our ability to uncover novel biology in areas where conventional approaches have struggled," Recursion CEO Najat Khan, PhD, said in a company statement quoted by GEN.

Winrow's remarks focus on how the map works—and why prior approaches missed what it surfaced.

"You need a certain differentiation period"

Winrow told GEN that cell context is non-negotiable. Recursion starts with human iPSCs and drives them into what he described as a "very clearly homogenous population of neurons" tested at scale.

"You need a certain differentiation period for the iPSCs to form into the neurons that you want to study," he said. "There's also a certain QC that you need to do to make sure that those neurons are what we expect them to be, and that that's robust and reproducible."

That scale is part of the story. The partners built what Recursion describes as the first whole-genome CRISPR knockout map from a subset of more than one trillion internally manufactured neuronal cells—roughly "12 brains' worth of neurons," according to GEN's reporting. Winrow said outside cell manufacturers initially told the partners the production volume was too difficult and too expensive.

Unbiased genome-wide scan, not a favorite pathway

The perturbation design is where Winrow draws the sharpest contrast with conventional neuroscience screens.

"We're looking at whole genome-wide knockout, not just a handful of areas or pathways of interest," he told GEN. "That's really a big game changer, in that we have this broad view. As we perturb the cells, we start to see the known actors. But more importantly, we're starting to see these unexplored areas really come to light."

Researchers applied whole-genome CRISPR-Cas9 knockouts spanning more than 17,000 genes alongside thousands of small molecules. Winrow described hunting for gene–compound and gene–gene interactions simultaneously.

Validation, he emphasized, is staged—not a single flashy hit call. Candidate targets moved through pathway validation, functional validation, and disease validation. Only targets with consistent evidence across stages entered the package Genentech accepted.

"Validation, then, is central to that core biology in exploring a whole bunch of different new intersecting targets," Winrow said.

Recursion says the path from initiating target validation to delivering the package took 15 months.

Forty-six million images—and what the AI sees that humans skim past

Imaging sits at the center of the platform narrative. Winrow told GEN the workflow captures more than 46 million cellular images, each analyzed across hundreds of morphological features—mitochondrial shape, nuclear morphology, and more.

"Traditionally as a scientist, I'd go in and I'm really interested in mitochondria," he said. "So, I look at the image and I say 'Wow, am I seeing mitochondrial fragmentation?' Well, that's great. Yes, I am, right?' But if you're looking at hundreds and hundreds of different features, that could be a mitochondrial shape, it could be nuclear capacity, it could be a whole bunch of different things that the AI models are trained upon, that then enables a real richness to come out of that dataset."

"So, it's not just the scale of those images, it's actually what's within those images that we require the AI approaches to really uncover," he added.

Downstream analysis runs on Recursion's BioHive-2 supercomputer, completed in 2024 on Nvidia hardware with proprietary models—turning phenotypic patterns into hypotheses that feed the validation pipeline.

Two CNS maps, six phenomaps, one commercial arc

The neuronal map is one of two neuroscience-focused whole-genome phenomaps in the Roche–Genentech partnership; the other maps microglial immune cells. Together they sit inside a broader set of six phenomaps, four of which target an undisclosed GI oncology indication.

Winrow highlighted reuse as a design advantage: insights from one map can be followed up in the other.

"Because we can reuse these maps again and again, I might find something in the microglia map that's really intriguing that I can now follow up in the neuron map, and vice versa," he told GEN.

Commercially, Recursion can receive up to $300 million in development and commercialization milestones per small-molecule program, plus tiered royalties up to high single digits. Roche and Genentech committed to advancing therapies across 40 programs spanning neuroscience and GI oncology.

What this does—and does not—prove

Winrow's public remarks describe a validated target, not an approved drug. Recursion's own cited statistics underscore how far discovery milestones sit from clinic success. The undisclosed target identity also means outside scientists cannot yet evaluate mechanism or tractability.

What the public record does support is a concrete contractual step in one of pharma's largest AI discovery alliances—and a detailed account of how phenomics at trillion-cell scale differs from pathway-first screening.

"Each target is individual," Winrow said. "They are related to a core disease process… It's an unbiased approach, so all of those things that are associated have not necessarily been connected to this target in the past, so you come up with all sorts of different targets."

For a field where most CNS targets recycle familiar biology, that claim is the story—until the chemistry platform produces a molecule worth naming.

### Sources

• GEN — Recursion Partners with Genentech to Advance First Validated Neuro Target Discovered Through AI Map (August 6, 2026)
• Recursion Pharmaceuticals — corporate statements as quoted in GEN coverage (August 2026)