The Challenge
Drug discovery is traditionally slow, expensive, and inefficient—especially for rare diseases, where the costs of experimentation at scale are often prohibitive. Recursion sought to accelerate this process using machine learning on high-resolution cellular images to detect drug-response patterns and identify promising therapeutic compounds. However, scaling such experimentation required not just advanced ML, but computational infrastructure capable of processing millions of data points weekly without bottlenecks or skyrocketing costs.
To realize its moonshot mission of “decoding biology to radically improve lives,” Recursion needed a faster, more cost-effective way to analyze and simulate drug treatments at scale.
The Solution
Tribe AI partnered with Recursion to improve the computational efficiency of their AI-driven platform by optimizing GPU cluster usage. By integrating supercomputing resources—including BioHive-1, one of the world’s most powerful supercomputers—with ML and computer vision models, the project focused on enhancing performance in the early stages of drug discovery.
The initiative delivered a 10x increase in computational efficiency, accelerating the pace of molecule screening and reducing both research timelines and costs for rare disease treatment.
Key Features
- ML-Enhanced Image Analysis: Leveraged machine learning to analyze high-resolution cellular images for drug-response indicators.
- Supercomputing Integration: Deployed the BioHive-1 cluster to run millions of experiments weekly—up to 2.2 million.
- GPU Optimization: Improved GPU cluster efficiency by 35%, unlocking significant cost savings.
- Failure-Tolerant Design: Allowed for large-scale hypothesis testing early in the funnel, embracing early-stage failure as a tool for learning.
- Cross-Functional Collaboration: Enabled effective coordination between biology, engineering, and data science teams through Recursion’s interdisciplinary leadership model.
How It Works
- Massive Experimentation Pipeline
Recursion’s system runs up to 2.2 million biological image-based experiments per week, capturing high-dimensional cellular data for analysis.
- Image-Based Phenotypic Modeling
ML models analyze cellular morphology to identify how compounds affect disease-relevant cell models, screening thousands of possibilities quickly.
- GPU & HPC Optimization
Tribe AI’s work focused on optimizing compute workloads to support these ML processes, improving GPU utilization and throughput on BioHive-1.
- Funnel Reinvention
Rather than narrowing drug candidates slowly, Recursion's platform flips the traditional “V” shaped funnel into a “T”—eliminating weak candidates early and allowing stronger leads to proceed with greater confidence.
Impact & The Future
The collaboration achieved measurable success:
- 35% improvement in GPU cluster efficiency
- $2.8M in annualized net value captured
- Accelerated research in high-priority therapeutic areas, especially rare diseases
- 10x increase in computational throughput
Beyond the technical win, this project helped operationalize Recursion’s strategy of pushing failure to the early stages of discovery—making experimentation faster, cheaper, and more scalable. Researchers now prioritize high-potential drug targets earlier, increasing confidence and success rates in downstream clinical trials.
Looking forward, Recursion’s AI-first, cross-disciplinary approach is positioned to drive even greater pharmaceutical innovation. Their scalable, supercomputer-powered ML platform sets a new benchmark for what’s possible in modern drug development—making the once impossible not only plausible, but practical.
