AIVIVE: A Novel AI Framework for Enhanced In Vitro to In Vivo Extrapolation (IVIVE) of Toxicogenomics Data

The dataset consists of transcriptomic profiles from rat liver tissue, curated from Open TG-GATEs database (link in references), along with predictions generated by the AIVIVE generator model.

The transcriptomic profiles are derived from both in vitro and in vivo experiments involving single-dose treatments of various compounds. The data is preprocessed using the RMA (Robust Multi-array Average) method, which ensures that the data is adjusted for batch effects and other systematic variations.

Training Data: 80% of the data is used for training the machine learning models. This subset is based on the unique compounds, meaning each compound has corresponding transcriptomic data across different exposures.

Test Data: 20% of the data is held back as a test set to evaluate the model's performance and generalization ability.

The dataset was obtained from Download - Open TG-GATEs | LSDB Archive. RMA normalization was performed in R (version 4.4.1). Additionally, the predictions from the optimal AIVIVE generator model for both training and testing sets are included that were used for further analysis.

Files:

vitro_train_test.csv: Train and test transcriptomic profiles from in vitro experiments

vivo_train_test.csv: Train and test transcriptomic profiles from in vivo experiments

generator1_encoded_prediction_9962160_VivoGenerator.csv: Train predictions from the optimal generator

generator1_encoded_prediction_9962160_vivoGenerator_test.csv: Test predictions from the optimal generator