GanCtrl: Synthetic Control Predictions for Liver and Kidney Clinical-Pathology Profiles (Open TG-GATEs)

This dataset contains real Open TG-GATEs–derived input files used to develop, train, and evaluate GanCtrl, a generative AI model that infers synthetic control profiles from time-matched treatment clinical-pathology data in repeated-dose rat studies from the Open TG-GATEs database. GanCtrl is trained to reconstruct the distribution of control responses conditional on treatment measurements, enabling reduction of real control animals in downstream analyses. The provided CSVs are real, preprocessed inputs derived from Open TG-GATEs studies and include: (1) clinical-pathology measurements, (2) study/sample metadata required for grouping and evaluation, and (3) molecular descriptor features used by GanCtrl. These files represent the time-matched treatment and control profiles consumed by GanCtrl for treatment→control translation, and they are formatted to support reproducibility of downstream analyses (e.g., cosine similarity, RMSE, and related agreement metrics) and exploration of alternative modeling strategies. Files: control_test_input: Real control input profiles for the held-out test split (with metadata + molecular descriptors). treatment_test_input: Real treatment input profiles for the held-out test split (with metadata + molecular descriptors). control_train_input: Real control input profiles for the training split (with metadata + molecular descriptors). treatment_train_input: Real treatment input profiles for the training split (with metadata + molecular descriptors). Note: This repository provides input files only; synthetic predictions and/or model checkpoints are distributed separately (see the GitHub link: CHANDMX20/GanCtrl.