ToxCML: Multi-End Point Toxicity Predictions for 54,601 Chemicals

ToxCML is a hybrid computational platform predicting 18 toxicity end points for 54,601 unique chemicals. The framework combines multiple molecular representations with machine-learning-based QSAR and read-across models, achieving reported AUC scores of approximately 0.86–0.99. The dataset was published by Fauzan Syarif Nursyafi on figshare in April 2026.

Use Cases

• Large-scale toxicity screening of new compounds based on the 18 predicted end points.
• Chemical hazard prioritization for regulatory assessment based on the hybrid mfCoQ-RASAR model predictions.
• Reducing reliance on animal testing by using the platform's in-silico predictions for initial safety evaluations.

Strengths

• Covers 54,601 unique chemicals, providing a broad chemical space for analysis.
• Reports high model performance with AUC approximately 0.86–0.99 and BACC 0.73–0.98 across end points.
• Incorporates tiered applicability-domain analysis, with in-domain coverage generally exceeding 95%.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is very small at 19.8 KB, suggesting it may contain only model metadata or summary results, not the full underlying chemical data.

Provenance

Source

figshare

Collection Method

Computational modeling integrating consensus QSAR and similarity-based read-across.

Time Range

null

Freshness

Last updated 2026-04-13 17:37:22; freshness should be verified.

Geography

null