AbstractThe majority of clinical trial failures are caused by low efficacy of investigated drugs, often due to a poor choice of target protein. Computational prioritization approaches aim to support target selection by ranking candidate targets in the context of a given disease. We propose a novel target prioritization approach, GuiltyTargets, which relies on deep network representation learning of a genome-wide protein-protein interaction network annotated with disease-specific differential gene expression and uses positive-unlabeled machine learning for candidate ranking. We evaluated our approach on six diseases of different types (cancer, metabolic, neurodegenerative) within a 10 times repeated 5-fold stratified cross-validation and achieved AUROC values between 0.92 - 0.94, significantly outperforming a previous approach, which relies on manually engineered topological features. Moreover, we showed that GuiltyTargets allows for target repositioning across related disease areas. Applying GuiltyTargets to Alzheimer’s disease resulted into a number of highly ranked candidates that are currently discussed as targets in the literature. Interestingly, one (COMT) is also the target of an approved drug (Tolcapone) for Parkinson’s disease, highlighting the potential for target repositioning of our method.AvailabilityThe GuiltyTargets Python package is available on PyPI and all code used for analysis can be found under the MIT License at https://github.com/GuiltyTargets.Author summaryMany drug candidates fail in clinical trials due to low efficacy. One of the reasons is the choice of the wrong target protein, i.e. perturbation of the protein does not effectively modulate the disease phenotype on a molecular level. In consequence many patients do not demonstrate a clear response to the drug candidate. Traditionally, targets are selected based on evidence from the literature and follow-up experiments. However, this process is very labor intensive and often biased by subjective choices. Computational tools could help a more rational and unbiased choice of target proteins and thus increase the chance of drug discovery programs. In this work we propose a novel machine learning based method for target candidate ranking. The method (GuiltyTargets) captures properties of known targets to learn a ranking of candidates. GuiltyTargets compares favorably against existing machine learning based target prioritization methods and allowed us to propose novel targets for Alzheimer’s disease.