EUROTOX 2026

Predictive Toxicology: Accelerating Off-Target Liability Mitigation through Physics-Based Cloud Workflows

Speaker:
Tatjana Braun, Senior Principal Scientist, Schrödinger

Abstract:
Unmanaged toxicity accounts for approximately 30%1 of drug discovery project failures. While the adoption of experimental screening panels has improved the safety profiles of drugs entering the market, the high cost and significant latency of these screens often delay their implementation until the later stages of pre-clinical discovery. Consequently, they cannot be effectively integrated into the hit-finding and lead-optimization phases. To address the need for early off-target screening, many teams utilize digital toxicology models based on ligand-based machine learning. Although these models are fast and cost-effective, they are frequently limited by poor generalizability to novel ligand structures and lack the necessary protein context to help designers rationally mitigate liabilities.

We present a novel, physics-based in-silico solution for identifying and mitigating off-target liabilities. This approach constructs a detailed 3D atomistic representation of the ligand-target interaction and utilizes free energy calculations to model off-target binding. Molecules can be assessed against individual targets or a screening panel of representative targets, with all calculations performed in the cloud to eliminate local hardware requirements. This workflow has been successfully applied to diverse protein classes. Here, we share both retrospective validation from literature and prospective applications in drug discovery projects, highlighting how this workflow effectively helped resolve toxicity-related liabilities in CYP3A4 and hERG across our internal pipeline.