AI Drug Discovery & Development Summit 2026

We're excited to be attending AIDDD in Boston. Don't miss our talk in the Discovery Chemistry track at 14:45 EST on October 28: A Recursive Closed-Loop System for New Computational Models and Approaches in Biopharma.

• Validating AI-generated hypotheses at scale is now the bottleneck.
• Our Nextflow-based closed-loop system tests thousands of hypotheses in parallel, guided by scientists, and tops public ADMET leaderboards.
• We'll share practical lessons on framing problems, measuring success, and what it takes to go from public benchmark success to solving problems in the lab.

Accurate models for ADMET, potency, selectivity, binding affinity and synthesisability form a key capability in bringing more molecules to the clinic safely. These better models can provide significant savings in time and downstream costs for the drug discovery and development process. Empirical validation of hypotheses is the foundation of progress in science. That matters even more now that AI agents can propose ideas faster than anyone can realistically test them. As computational capability grows, execution and validation at scale become critical.

We built a closed-loop discovery system using Nextflow that turns every hypothesis into a reproducible pipeline, runs thousands in parallel, scores each on held-out data, learns and plans the next round. It is agnostic to the scientific problem: anything with data and a measure of success can go through the same loop, and scientists steer it with their own domain expertise.

On public ADMET benchmarks, from metabolic clearance and plasma protein binding to CYP metabolism and hERG liability, it produces small, task-specific ML models that top the leaderboards. In this presentation we'll show what it takes in practice: framing the question, building accurate measures of success, how scientists' deep domain knowledge steers the search, and the differences between public benchmarks and partnership projects inside pharma.