ADMET Predictions Get AI Boost, Federated Data Network Unites Pharma

Why It Matters

By delivering more accurate ADMET predictions early in discovery, the network can slash costly late‑stage failures and speed time‑to‑market, while demonstrating a scalable, privacy‑preserving AI collaboration model for the pharma industry.

Key Takeaways

• Federated ADMET model trained on 80% data from five firms
• Secure local inference keeps proprietary data within each company's environment
• Network aims to cut 90% failure rate via better ADMET
• Open to new members, continuously expanding shared chemical space
• Planned expansion to PROTACs, peptides, macrocycles beyond small molecules

Pulse Analysis

The high attrition rate in drug development is largely driven by poor ADMET properties, which can derail a candidate late in the pipeline and inflate R&D costs. Traditional approaches rely on isolated datasets, limiting model robustness across diverse chemical spaces. Federated learning offers a solution by aggregating insights from multiple organizations while keeping raw data behind each company’s firewall, thereby reconciling the need for data volume with stringent IP protection.

Apheris’s ADMET Network operationalizes this concept by creating a shared foundation model trained on the majority of data contributed by its five founding pharma partners. Participants can locally fine‑tune the model to reflect their specific pipelines, ensuring predictions are both globally informed and program‑relevant. Compared with Eli Lilly’s TuneLab, which primarily leverages Lilly’s internal data, the ADMET Network’s governance structure is designed for large‑scale, multi‑company collaboration, reducing strategic dependencies while enhancing model generalizability across the broader chemical universe.

The broader industry impact is significant. As more firms join, the collective chemical coverage expands, accelerating the discovery of viable molecules and shortening synthesis cycles. Future extensions to PROTACs, peptides, and macrocycles promise to democratize AI benefits beyond small‑molecule drugs. Ultimately, this federated ecosystem could reshape R&D economics, delivering faster, safer therapeutics while preserving competitive advantage—a compelling blueprint for AI‑driven collaboration in biopharma.