The evolution of next-generation sequencing technologies has facilitated the detection of causal genetic variants in diseases previously undiagnosed at a molecular level. However, in genome sequencing studies, the identification of disease genes among a candidate gene list is often difficult because of the large number of apparently damaging (but usually neutral) variants. A number of variant prioritization tools have been developed to help detect disease-causal sites. However, the results may be misleading as many variants scored as damaging by these tools are often tolerated, and there are inconsistencies in prediction results among the different variant-level prediction tools. Recently, studies have indicated that understanding gene properties might improve detection of genes liable to have associated disease variation and that this information improves molecular diagnostics. The purpose of this systematic review is to evaluate how understanding gene-specific properties might improve filtering strategies in clinical sequence data to prioritize potential disease variants. Improved understanding of the \'disease genome\', which includes coding, noncoding and regulatory variation, might help resolve difficult cases. This review provides a comprehensive assessment of existing gene-level approaches, the relationships between measures of gene-pathogenicity and how use of these prediction tools can be developed for molecular diagnostics.