Read activity in a specific assay.
Labels indicate activity in named biological assays. The 0.5 threshold is exploratory; neural scores have not been calibrated as probabilities of harm.
Predictions compared with published Tox21 measurements. The outcomes, the errors, and the conditions that shape the result.
Read the test reportRetrospective experimental assay evidence.
Not a clinical or chemical safety validation.
Exactly 1,000 distinct stereo-free fragment parent structures. The 883 external-source compounds are reported separately in the PDF. Reference labels are measured assay outcomes; missing observations are excluded per endpoint.
Each assay contributes equally to the macro average. ROC-AUC measures ranking; average precision emphasizes retrieval of active assay observations.
Higher is better for ranking. This is not percentage accuracy.
| Model | ROC-AUC | Average precision | 95% bootstrap AUC interval |
|---|---|---|---|
| GCNGraph neural network | 0.780 | 0.280 | [0.754, 0.803] |
| GINEGraph neural network | 0.777 | 0.258 | [0.749, 0.802] |
| GATv2Graph neural network | 0.755 | 0.274 | [0.725, 0.783] |
| ECFP-RFFingerprint baseline | 0.737 | 0.269 | [0.709, 0.765] |
| ECFP-LRFingerprint baseline | 0.686 | 0.250 | [0.651, 0.720] |
All 1,000 molecules also passed through the application API in ten sequential batches.
Model–molecule parity comparisons
Service checks passed
Maximum absolute score difference
Validation study / 2026Assay-level tables, ROC and precision-recall curves, real measured examples, calibration, uncertainty and model disagreement.
Labels indicate activity in named biological assays. The 0.5 threshold is exploratory; neural scores have not been calibrated as probabilities of harm.
Bootstrap intervals describe uncertainty for the tested weights and compounds. They do not measure variability across independent neural training runs. Chemical dependence and scaffold overlap can affect them.
The cohort manifest preserves source IDs, structures and observed labels. External-only results remain separate from the historical held-out subset.
NIH Tox21 Challenge data