Uncertainty-Quantified QSAR Predictions: VEGA Models (98 Endpoints) with Conformal Intervals/Sets for Training, Test, and EPA CompTox Datasets (500K Compounds)

# Dataset Description This repository contains training data, test data, and uncertainty-quantified predictions for 98 QSAR models from the VEGA platform, plus large-scale screening results for the EPA CompTox database. ## Contents ### 1. VEGA Training and Test Sets- **51 regression endpoints** - **52 classification endpoints** - ~114,000 predictions across VEGA test sets- Molecular structures (SMILES), experimental values, model predictions- Applicability Domain Index (ADI) for each prediction ### 2. Conformal Prediction Results (VEGA)- **Prediction intervals** for regression (90% coverage)- **Prediction sets** for classification (90% coverage) - Auxiliary model confidence scores- Conformal prediction metrics ### 3. VEGA point prediction and Conformal Prediction Results (EPA CompTox ) - **Prediction intervals** for regression (90% coverage)- **Prediction sets** for classification (90% coverage) - Auxiliary model confidence scores- Conformal prediction metrics - ADI values for all predictions This is a supplementay information for a paper submitted to https://pubs.acs.org/doi/10.1021/acs.chemrestox.5c00293 Update: Manuscript revision version 2026-05-16 Excel files for VEGA QSAR Training and Test Sets now have empty sheet if the test set is missing (was a duplicated training set) Classification results now have proper set size and singleton efficiency (due to fixed bug in classification pipeline) Added VEGA predictions for training/test datasets and EPA CompTox subset in original VEGA format in order to help reproducing the pipeline Code to reproduce the results is at https://github.com/ideaconsult/quarkus-vega-cli and https://github.com/ideaconsult/qubounds