The computational prediction of compound effects from molecular data is an important task in hazard and risk assessment and pivotal for judging the safety of any drug, chemical or cosmetic compound. In particular, the identification of such compound effects at the level of molecular interaction networks can be helpful for the construction of adverse outcome pathways (AOPs). AOPs emerged as a guiding concept for toxicity prediction, because of the inherent mechanistic information of such networks. In fact, integrating molecular interactions in transcriptome analysis and observing expression changes in closely interacting genes might allow identifying the key molecular initiating events of compound toxicity.In this work we describe a computational approach that is suitable for the identification of such network modules from transcriptomics data, which is the major molecular readout of toxicogenomics studies. The approach is composed of different tools (1) for primary data analysis, i.e., the biostatistical quantification of the gene expression changes, (2) for functional annotation and prioritization of genes using literature mining, as well as (3) for the construction of an interaction network that consists of interactions with high confidence and the identification of predictive modules from these networks. We describe the different steps of the approach and demonstrate its performance with public data on drugs that induce hepatic and cardiac toxicity.
Keywords: Computational modeling; Drug Toxicity; Literature mining; Molecular interactions; Network analysis; Toxicogenomics.