Clustering chemicals based on similarities in structure, properties, or modes of action aids in predicting the behavior, properties, and potential hazards of chemicals, particularly when combined with toxicity bioactivity data. Chemical clusters, when combined with toxicological mechanisms and methods such as structure-activity relationships, can be used as part of integrated or tiered testing strategies.

The Endocrine Disruptor Screening Program (EDSP) Universe of Chemicals (UoC) describes a list of chemicals to be screened for endocrine disruption. Understanding the coverage of test methods for the UoC is important in guiding data interpretation. This study clustered 6,952 chemicals from the EDSP UoC with identifiable chemical structures using SMILES-based ToxPrint fingerprints into 826 clusters. Subsequently, ToxCast/Tox21 chemicals were mapped to the EDSP clusters to investigate the overlap in chemical space. Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction was used to visualize overlaps between the datasets. Overall, there is a large overlap in structural space between the chemicals in the EDSP UoC and ToxCast/Tox21 with ~40% of chemicals present in both datasets. Even so, 14 of the 489 UMAP grids (containing 18 UoC chemicals) highlight areas of the EDSP UoC not covered by comparable chemicals tested in ToxCast/Tox21. Similar observations were made when using physicochemical properties, with 6 of the 346 UMAP grids containing only EDSP UoC chemicals (42 chemicals) and no ToxCast/Tox21 chemicals. These areas highlight groupings where further testing is needed to improve extendibility of the ToxCast/Tox21 to the of EDSP UoC space.

This abstract does not necessarily reflect U.S. EPA policy.