To develop a new model for predicting compound toxicity and exploring related toxicity mechanisms using transcriptomic data and gene ontology knowledge.

Using the TOXRIC database, two toxicity-related datasets were constructed and a Tox VNN model was established that incorporated gene ontology knowledge to evaluate compound toxicity and identify key biological processes.

Tox VNN demonstrated good predictability. The identification of important biological processes related to CYP enzyme activity and p53 pathway stress response provided insights into the toxicity mechanisms.

The Tox VNN, which integrates data and knowledge, can not only ensure high predictability, but also effectively identify important biological processes related to toxicity. This model offers a new approach to predicting and understanding compound toxicity in drug safety evaluation.