Particle.news

Lilly Launches TuneLab, Opening Proprietary AI Drug-Discovery Models to Startups

Participants contribute data through a federated system that lets Lilly improve its models without exposing proprietary information.

Overview

  • TuneLab launches with 18 models: 12 for small-molecule property prediction and six for antibody developability.
  • Training runs through Rhino Federated Computing using Nvidia technology so models learn from partner data without exposing proprietary datasets.
  • About a dozen startups have joined, including Insitro, Circle Pharma, Firefly Bio and Superluminal Medicines, with Insitro set to build additional models for small-molecule discovery.
  • The models draw on years of Lilly experiments, supported by more than $1 billion in investment and data from hundreds of thousands of unique molecules.
  • Participation flows through Lilly’s Catalyze360 program, which offers capital, lab space, technology and expertise, and is limited to biotechs already in preclinical development with a small-molecule or antibody candidate.