BIOLOGY.DIGITAL
Back to Case Studies

Roche & Genentech + Recursion Pharmaceuticals: Whole-Genome Phenomic Maps in Neuroscience & GI Oncology

Pharmaceutical R&D100,000+ employees ($65B+ annual revenue)Recursion Pharmaceuticals

📌Key Takeaways

  • 1Roche & Genentech (Pharmaceutical R&D, 100,000+ employees ($65B+ annual revenue)) deployed Recursion Pharmaceuticals.
  • 2Upfront & Milestone Capital Validated: $180M realized (now $180M ($150M upfront + $30M phenomap acceptance)).
  • 3Weekly High-Content Screening Throughput: 44x throughput (now 2.2 million automated wells/week).
  • 4Implementation timeline: 24 months to first validated whole-genome neuroscience phenomap ($30M milestone).
**Deployment Snapshot:** - Partner: Roche & Genentech - Platform you're evaluating: Recursion Pharmaceuticals - Core domain: AI Drug Discovery - Why we track this deployment for our benchmarks: Verified peer-reviewed and public filing milestones

Architectural Context & Scientific Rationale

Single-target hypothesis-driven discovery struggles in complex polygenic neurological disorders and heterogeneous gastrointestinal malignancies where compensatory signaling networks blunt single-pathway blockade. In December 2021, Roche and Genentech entered a multi-year collaboration with Recursion Pharmaceuticals to replace linear target validation with industrialized high-content phenomic mapping. Rather than pre-selecting candidate proteins from literature, the partnership Perturbs every gene in the human genome via automated CRISPR-Cas9 knockouts alongside hundreds of thousands of small molecules in human induced pluripotent stem cell (iPSC)-derived neurons and gastrointestinal primary cells. What's the takeaway for your own architecture? We've analyzed this rollout closely, and it's clear you can't separate compute from wet-lab execution. You'll get the best ROI when your scientists and modelers share one data loop.

Wet-Lab Automation & Supercomputing Execution

Recursion's Salt Lake City robotic wet labs execute up to 2.2 million multi-channel fluorescence microscopy and pooled transcriptomic assays weekly. Vision transformer foundation models trained on the NVIDIA-powered BioHive-2 supercomputer (504 H100 GPUs) convert raw 6-channel Cell Painting organelle images into mathematical embeddings that quantify subtle morphological shifts across billions of single cells. In September 2024, Roche and Genentech formally exercised their option to accept the first whole-genome neuroscience phenomap, triggering a $30 million milestone payment to Recursion and advancing multiple novel target hypotheses into hit-to-lead optimization. What's the takeaway for your own architecture? We've analyzed this rollout closely, and it's clear you can't separate compute from wet-lab execution. You'll get the best ROI when your scientists and modelers share one data loop.

Upfront & Milestone Capital Validated

$180M realized

Weekly High-Content Screening Throughput

44x throughput

Proprietary Phenomic & Multi-Omic Dataset

60+ PB unified

The Challenge

Conventional literature-driven target discovery in neurodegenerative diseases and GI oncology suffers high Phase II attrition because animal models and bulk transcriptomics fail to capture human cell-type-specific network compensation. That's a massive bottleneck for any R&D team. If you're running similar programs, you know legacy workflows simply don't scale.

The Solution

Deployed Recursion OS robotic wet labs and vision foundation models on BioHive-2 to map whole-genome CRISPR knockouts and chemical library perturbations across human iPSC-derived neuronal and GI cells. Here's how they fixed it. By pairing Recursion Pharmaceuticals with automated wet-lab feedback, we see how you can compress months of manual screening into weeks.

Implementation

24 months to first validated whole-genome neuroscience phenomap ($30M milestone)

Joint Genentech Research & Early Development (gRED), Roche pRED, and Recursion computational phenomics teams

  1. 1Differentiated human iPSC lines into reproducible cortical neuronal and gastrointestinal cellular models compatible with 1536-well robotic liquid handling.
  2. 2Executed whole-genome CRISPR-Cas9 knockouts and chemical library screens using standardized 6-channel Cell Painting fluorescence microscopy.
  3. 3Trained vision transformer foundation models on BioHive-2 (504 NVIDIA H100 GPUs) to embed single-cell morphology into a queryable mathematical covariance matrix.
  4. 4Delivered the first whole-genome neuroscience phenomap to Roche/Genentech, triggering formal acceptance and a $30M milestone payment.

Results

MetricBeforeAfterChange
Upfront & Milestone Capital ValidatedPre-partnership baseline$180M ($150M upfront + $30M phenomap acceptance)$180M realized
Weekly High-Content Screening Throughput<50,000 manual assay wells/week2.2 million automated wells/week44x throughput
Proprietary Phenomic & Multi-Omic DatasetIsolated project datasets60+ Petabytes relatable across human cell types60+ PB unified
"This collaboration highlights the potential of technology to transform drug discovery and get previously unknown insights into complex disease in an unbiased way."

Dr. James Sabry, Former Global Head of Pharma Partnering, Roche

Key Learnings

  • 1Industrialized phenotypic screening requires strict reagent lot control and robotic environmental standardization so batch effects do not masquerade as biological signals.
  • 2Pairing genetic CRISPR knockouts with small-molecule perturbations in the same cell line enables bidirectional mapping between novel targets and druggable chemotypes.
  • 3Large-scale supercomputing (500+ H100 GPUs) is mandatory when training self-supervised vision transformers on billions of raw single-cell microscopy crops.

Frequently Asked Questions

In September 2024, Roche and Genentech accepted the first whole-genome neuroscience phenomap generated on Recursion OS, paying a $30 million milestone fee after verifying causal genetic-to-chemical relationships in human iPSC-derived neuronal cell types. If you're negotiating a similar partnership, we recommend benchmarking your internal assay throughput first.
The collaboration covers up to 40 neuroscience and gastrointestinal oncology programs, with $150 million paid upfront and up to $300M to $500M in research, development, and commercialization milestones per program. If you're negotiating a similar partnership, we recommend benchmarking your internal assay throughput first.
Learn More About Recursion Pharmaceuticals

Verified References & Citations (2)

  1. [1]

    Roche and Genentech accept first neuroscience phenomap from Recursion, triggering $30M milestone payment.

  2. [2]

    Recursion and Roche-Genentech collaboration scope across 40 programs in neuroscience and GI oncology.

Last updated: March 1, 2026

Ask AI