Predicting the effects of a treatment with a new compound is a fundamental challenge in biomedicine and includes the identification of both the compound's target and off-target effects. For two decades, computational tools have been proposed to address this challenge, leveraging prior information in the form of known compound-target relationships. Available “guilty-by-association” solutions can be grouped into three broad categories: chemoinformatics-based approaches that use chemical structure similarities, text-mining approaches (see read-across) that dig into literature information to identify previously reported observations, and approaches that cross-match experimental outcomes with the above. In this review, we introduce the main components that make systems toxicology a powerful approach to predict the likelihood of undesired drug effects and provide insights into their mechanisms. The integration of high-throughput technologies and advanced computational approaches that employ complex network models with standard endpoints from pre-clinical and clinical studies allows system wide evaluation of the effects a drug can elicit on the biological system, including the adverse ones.