Bridging Organ Transcriptomics for Advancing Multiple Organ Toxicity Assessment with a Generative AI Approach

Abstract

Translational research in toxicology is essential for understanding how molecular alterations manifest across various biological systems to, for example, decrease reliance on animal models and extrapolation from animals to humans. Toxicogenomics (TGx) significantly contributes to assessing chemical and drug toxicity by providing insights into underlying toxicity mechanisms and developing gene expression-based biomarkers for toxicant classification. Despite the recognized need for a multi-organ approach in evaluating organism-level toxicity, most TGx research has been focused on a limited number of organs, primarily the liver, due to resource-intensive experiments. This paper is the first effort to utilize Generative Adversarial Network (GAN) for bidirectional translation of transcriptomic profiles between organs under chemical treatment. In this study, we developed a novel GAN model, TransTox, to bridge transcriptomic data between the liver and kidney. This model demonstrated robust performance in various evaluations, including external validation on independent datasets from both the training set’s source labs and a different lab. The study investigated the concordance between the real data and synthetic data generated by TransTox in elucidating toxicity mechanisms with respect to differential expressed genes (DEGs) and enriched pathways analyses. It showed comparable results in comparison to that obtained from real experimental settings. Moreover, TransTox proved valuable in biomarker applications, where synthetic data could be used to develop valid biomarkers or serve as “digital twins” for diagnostic applications. TransTox holds the potential to extend insights into toxicological effects in other organs by leveraging historical liver-centric TGx experiments, thereby opening avenues for reducing reliance on animal testing in organ toxicity research.