What if human clinical trials were predictable?

Our virtual human models generate human biology. Use VERDICT to evaluate assets, investments, and trial designs.

Virtual human models. Future human experiments.

GEM is our family of generative expression models. It learns human biology to predict drug response before anyone runs the experiment.

Our Science

Trained on human biology.

Our models learn directly from one of the world’s largest harmonized human RNA-seq collections, connecting biological measurements with curated experimental context and clinical trial outcomes.

bulk RNA-seq samples: 800,000
single cells: 400M
unique perturbation contexts: 22,300
human tissues: 50+
therapeutic areas: 40+
clinical trials with labelled outcomes: 12,418

From generated biology to clinical predictions.

VERDICT uses predicted human biology to forecast trial outcomes, helping you compare treatment effects and safety before running a study.

Significance: Will the primary analysis meet its statistical test?
Direction: If it does, which arm does it favor?
Effect size: How large is the treatment effect?
Adverse events: What serious adverse events are expected in each arm?

What you can do with VERDICT

Frontier trial prediction

35%

Relative increase in observed success rate.

Among trials VERDICT called successful, compared with all evaluated trials.

In retrospective testing on 1,717 held-out later trials.

Success means at least one significant primary endpoint, not approval.

See the benchmarks

Observed success rate

All evaluated trials

52%

Trials VERDICT called successful

70%

What can you do with VERDICT?

Predict a future trial

Enter a trial design or registered trial ID. VERDICT combines the design with predicted human biology to forecast treatment effects and safety, with uncertainty to guide your decision.

Try VERDICT

Enter trial design or NCT IDGenerate human biologySignificance, direction, effect, safety

Optimize a future trial

Compare populations, comparators, endpoints, and doses. Use predicted outcomes to choose a trial design before enrolling patients.

Try VERDICT

One assetDesign, predict, reviseCompare predicted outcomes

Optimize an asset

Evaluate an asset across indications, doses, and combinations. Compare predicted treatment effects and safety to prioritize the most promising development paths.

Try VERDICT

Fix a trial designPredict, modify, repeatOptimized asset

Improve a model

Use GEM-generated biological responses and VERDICT trial predictions as feedback for AI models. Evaluate proposed assets and guide the next round of development.

Biology in the learning loop

LLM promptGenerate, reward, optimizeBetter biological reasoning

Built by experts in AI, biology and clinical development.

Jeff Leek, PhD· TIME100 AI 2025
Co-Founder, CEO
A TIME100 AI honoree and winner of the Presidents’ Award for top statistician under 40, whose open-source genomics software has been downloaded more than 1.4 million times.
Robert Bradley, PhD
Co-Founder, SAB Chair
A Fred Hutch computational biologist whose research showed how faulty RNA splicing drives cancer, and a co-founder of Codify Therapeutics, which develops programmable RNA-targeted therapies.

11 PhD scientists from BMS, J&J, Recursion, Novo, Pathos, Freenome, Genentech

Meet the team