The problem at the heart of drug discovery: Lexogen & Ochre Bio on the power of AI on human data

What It Covers

Ochre Bio CEO Quinn Wills and Lexogen CEO Stefan Baj examine why drug discovery fails due to non-human biology models, how they built one of the world's largest human liver functional genomics datasets, and how purpose-designed RNA sequencing data enables AI-driven target discovery for liver disease affecting 1.5 million deaths annually.

Key Questions Answered

• •Human data gap in liver disease: Only 40,000 liver transplants are performed globally each year against 1.5 million annual deaths from liver disease. Drug discovery fails here across three compounding problems: insufficient causal human biology knowledge, animal models that cannot predict fibrosis regression, and clinical trials that lack reliable human endpoint data — all requiring a human-first data strategy to address.
• •AI requires causal data, not just correlational data: AI models detect correlations that human brains cannot see, but correlation alone cannot resolve biological complexity or causality. To build genuinely predictive models, teams must generate perturbation data — knocking out every expressed gene across multiple human donors and disease states — then RNA sequence the results to construct a functional gene regulatory network.
• •Designing data for AI at scale: Ochre Bio and Lexogen generated a foundational perturbation dataset covering all expressed genes across multiple human liver donors and disease states. Lexogen's TAW technology, which eliminates the RNA extraction step, reduced variation and increased sensitivity, enabling detection of low-level gene expression. The project achieved a 95% success rate and was completed within five months.
• •Lab-in-the-loop scaling strategy: Ochre Bio perturbs approximately 2,000 genes directly in complex human liver tissue models, then trains AI on that causal data to predict outcomes for all remaining untested genes in silico. This approach avoids reverting to simple cell lines while scaling predictions beyond what physical experimentation alone can achieve, using complex human tissue as the AI training ground.
• •Regulatory environment shifting toward human models: The FDA no longer requires animal studies before first-in-human trials, and recently indicated that liver disease clinical trials can move away from biopsy-dependent endpoints toward biomarker-driven measures. Bayesian clinical trial designs now allow data borrowing across prior studies and indications, enabling more sophisticated and ethically grounded use of human data in drug development.