Published September 18, 2026 | Version v1

TransHazard

Authors/Creators

Description

TransHazard is a reproducibility package supporting the study “TransHazard: Directionally Constrained Modeling of Ecotoxicity Inheritance in Environmental Transformation Products.” The framework was developed to evaluate whether a transformation product becomes more toxic, less toxic, or remains comparable with its parent chemical using paired parent-to-product ecotoxicity evidence. Transformation relationships were integrated with biologically comparable EPA ECOTOX records, converted to a signed hazard-difference target (ΔH), and evaluated using leakage-controlled, component-disjoint validation.

The frozen benchmark contains 216 non-self-loop parent–transformation-product edges organized into 83 connected chemical components and a predefined three-fold validation split (locked_component_3fold_v2). The archive includes model-ready datasets permitted for redistribution, frozen split assignments, categorical and continuous out-of-fold predictions, validation and sensitivity-analysis outputs, applicability-domain results, Y-scrambling and component-bootstrap analyses, figure source-data tables, publication tables, workflow manifests, audit files, and executable analysis scripts.

The primary categorical model is a hierarchical directionally constrained classifier for MORE_TOXIC, LESS_TOXIC, and COMPARABLE outcomes, while the complementary continuous model is an exact antisymmetric regression model for estimating signed ΔH. The repository is intended to support transparent reproduction, auditing, and extension of the computational workflow.

Raw third-party source records from the U.S. EPA ECOTOX Knowledgebase, EPA Chemical Transformations Database (CheT), and NORMAN SusDat/Suspect List Exchange are not redistributed in this archive. These resources should be accessed directly from the respective source providers and cited according to their current terms of use.

Files

TransHazard_Zenodo_Package_Builder.zip

Files (303.1 kB)

Name Size
md5:83bfbda9c8c13c45d23000cfe2e9d470
303.1 kB Preview Download