BioTechX 2026

Predictive Toxicology: Rational Digital Toxicology in the Cloud with a New AI-Accelerated Physics-Based Workflow

Speaker:
Steven Jerome, Executive Director, Life Science Software, Schrödinger

Abstract:
By one estimate, unmanaged toxicity is responsible for roughly 30%1 of all drug discovery project failures. The adoption of experimental screening panels have contributed to the overall improved safety profile of drugs on the market. However, the high cost and latency associated with performing these screens means that such panels are run later in the pre-clinical discovery process and cannot be effectively incorporated into hit finding and lead-optimization stages of the project. To meet the demand for off-target screening during the design process, many teams deploy digital toxicology screening in the form of ligand-based machine learning models. These models, which are fast and inexpensive to operate are typically limited by poor generalizability to ligand matter dissimilar from data used to train the models and are missing the protein context to help designers dial-out liabilities rationally. We present a novel in-silico, physics-based solution for the identification and mitigation of off-target liabilities that constructs a full 3D, atomistic, representation of the ligand interacting with the target and leverages free energy calculations to model off-target binding. Molecules can be evaluated against a single off-target or a panel of representative targets in a screening mode. Calculations are run in the cloud, eliminating any need for local hardware. AI and ML models trained to the physics-based predictions have significant potential to enable high-throughput application in the near future. Already, this workflow has been successfully applied to a wide range of relevant targets across many protein classes. Here, we present both retrospective validation from literature data and prospective application to internal drug discovery projects, where the workflow has seen significant impact throughout our internal drug discovery pipeline, emphasizing the efficient resolution of tox-related liabilities in CYP3A4 and hERG.