An ensemble learning approach for modeling the systems biology of drug-induced injury

Abstract Background Drug-induced liver injury (DILI) is an adverse reaction caused by the intake of drugs common use that produces damage. The impact DILI estimated to affect around 20 in 100,000 inhabitants worldwide each year. Despite being one main causes failure, pathophysiology and mechanisms are poorly understood. In present study, we developed ensemble learning approach based on different features (CMap gene expression, chemical structures, drug targets) predict might cause gain a better understanding linked reaction. Results We searched for signatures CMap expression data using two approaches: phenotype-gene associations from DisGeNET, non-parametric test comparing DILI-Concern No-DILI-Concern (as per DILIrank definitions). average accuracy classifiers both approaches was 69%. used structures as features, obtaining 65%. combination types produced 63%, but improved independent hold-out up 67%. drug-target feature obtained best (70%) test. Conclusions When data, searching specific signature among landmark genes improves quality classifiers, it still limited intrinsic noise dataset. feature, structural diversity known DILI-causing hampers prediction, which similar problem information. did not improve increased robustness shown tests. specially specificity, results were comparable previous research studies.