Discriminating the causative disease variant(s)for individuals with inherited or de novo mutationspresents one of the main challenges faced by theclinical genetics community today. Computationalapproaches for variant prioritization includemachine learning methods utilizing a large number offeatures, including molecular information,interaction networks, or phenotypes. Here, wedemonstrate the PhenomeNET Variant Predictor (PVP)system that exploits semantic technologies andautomated reasoning over genotype-phenotyperelations to filter and prioritize variants in wholeexome and whole genome sequencing datasets. Wedemonstrate the performance of PVP in identifyingcausative variants on a large number of syntheticwhole exome and whole genome sequences, covering awide range of diseases and syndromes. In aretrospective study, we further illustrate theapplication of PVP for the interpretation of wholeexome sequencing data in patients suffering fromcongenital hypothyroidism. We find that PVPaccurately identifies causative variants in wholeexome and whole genome sequencing datasets andprovides a powerful resource for the discovery ofcausal variants.