Generating a Novel Dataset for Mechanisms of Drug-Induced Toxicity using LLM-supported tools

Sara Masarone, Alex Beeson, Katie Victoria Beckwith, Matthew Mason, Thomas Clelford, Arran Willmott, Giovanni Montana, Layla Hosseini-Gerami

Confirmation: I have read and agree with the workshop's policy on behalf of myself and my co-authors.

Track: tiny / short paper (2-4 pages excluding references; extended abstract format)

Keywords: LLM, drug-discovery, applied ML, benchmarks, data integration

TL;DR: ToxMech is an LLM-supported system that integrates diverse data sources into a heterogeneous knowledge graph to map the causal mechanisms of drug-induced toxicity for improved safety assessment.

Abstract:

Toxicity is a leading cause of drug failure, yet existing resources often lack the mechanistic context linking drug perturbations to adverse outcomes. To bridge this gap, we introduce ToxMech, an ongoing project developing an LLM-supported system that extracts and structures toxicity mechanisms into a comprehensive heterogeneous knowledge graph.

ToxMech integrates data from diverse sources, including PubMed, AOP-Wiki, FDA boxed warnings, and clinical news, using retrieval-augmented agents to mine both structured repositories and unstructured text. By encoding relationships between drugs, targets, pathways, and outcomes, ToxMech enables structured reasoning over the causal chains of drug-induced toxicity. This evolving resource aims to provide researchers with a robust tool for mechanistic modelling and enhanced safety assessment in the drug discovery pipeline.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Presenter: ~Sara_Masarone2

Format: Yes, the presenting author will attend in person if this work is accepted to the workshop.

Funding: No, the presenting author of this submission does *not* fall under ICLR’s funding aims, or has sufficient alternate funding.

Submission Number: 88