Bridging organ transcriptomics for advancing multiple organ toxicity assessment with a generative AI approach

Area of Science:

• Toxicology
• Computational Biology
• Genomics

Background:

• Transcriptomic profiling is crucial for drug safety assessment but often limited to the liver due to resource constraints.
• Developing methods for multi-organ transcriptomic analysis is essential for comprehensive drug safety evaluation.

Purpose of the Study:

• To develop an AI model, TransTox, for bidirectional translation of transcriptomic profiles between liver and kidney under drug treatment.
• To validate TransTox's performance using independent datasets and assess its utility in characterizing toxicity mechanisms and predictive modeling.

Main Methods:

• Utilized a generative adversarial network (GAN) to create the TransTox AI model.
• Performed bidirectional translation of transcriptomic data between liver and kidney.
• Validated the model against experimental data and assessed its performance in gene expression predictive tasks.

Main Results:

• TransTox successfully generated synthetic transcriptomic data that accurately characterized toxicity mechanisms.
• Synthetic data from TransTox proved valuable for developing gene expression predictive models, acting as digital twins.
• The model demonstrated robust performance across independent datasets and laboratories.

Conclusions:

• TransTox facilitates AI-driven multi-organ toxicity assessment by enabling transcriptomic data translation.
• The AI approach holds significant potential for advancing precision toxicology and drug safety evaluation.
• TransTox offers a novel solution to overcome resource limitations in organ-specific transcriptomic studies.