Predicted NOEC Toxicity for 10,863 PFAS Across 13 Fish Species

Over 10,000 per- and polyfluoroalkyl substances (PFAS) have predicted no-observed-effect concentrations (NOECs) for fish mortality, generated by a machine learning model. The model, developed by Minhao Wang and uploaded in May 2026, used molecular structural descriptors and achieved a test R² of 0.7096. Predictions were extrapolated to 12 additional fish species using trait-based inference, revealing toxicity variations linked to chemical chain length and species traits like diet and temperature preference.

Use Cases

• Prioritizing PFAS for regulatory review based on predicted toxicity profiles for Danio rerio.
• Investigating structure-activity relationships for PFAS toxicity based on molecular descriptors like chain length and polar functional groups.
• Assessing trait-mediated differences in chemical sensitivity across fish species, such as warm water versus cold water or omnivorous versus carnivorous species.
• Screening large chemical inventories for ecological risk using the described machine learning and extrapolation framework.

Strengths

• Contains predicted NOECs for 10,863 distinct PFAS chemicals, a large-scale prediction for a class of over 10,000 persistent substances.
• Model performance is quantified with a specific test R² value of 0.7096, and predicted NOEC ranges are provided (0.239 to 163.452 mg/L).
• Toxicity profiles are extrapolated to 12 additional fish species beyond the primary model species (Danio rerio), enabling cross-species analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific analytical tasks.
• The dataset is small (56.0 KB), indicating a limited scope likely focused on summary predictions rather than raw experimental data.

Provenance

Source

Minhao Wang via figshare.

Collection Method

Machine learning predictions using compiled toxicity data and molecular structural descriptors, with trait-based inference for species extrapolation.

Freshness

Last updated 2026-05-21 09:22:34; freshness should be verified.