Skip to main content
Fig. 3 | Genome Medicine

Fig. 3

From: A visual and curatorial approach to clinical variant prioritization and disease gene discovery in genome-wide diagnostics

Fig. 3

Solved and unsolved cases in the BMGL cohort. In 245 exome cases, 51 had reported molecular diagnoses. The solved cases tended to have (a) more Human Phenotype Ontology (HPO) phenotypes, including ontological parent terms (Wilcoxon p = 0.0302); b higher average similarity to the Online Mendelian Inheritance in Man (OMIM) catalog (Resnik similarity, Wilcoxon p = 0.0387); and (c) lower quantities of filtered variant (Wilcoxon p = 0.2177). d Visualization. Multi-dimensional scaling representation of the 51 solved (yellow spheres) and 194 unsolved (red spheres) cohort cases in a three-dimensional map of all 7,746 cataloged OMIM diseases (gray spheres). Solved and unsolved cases appear similarly distributed in the visual space. e Transitive method comparison. Across the 47 solved cases with reported Morbidmap genes, we tested maximum, mean, and sum as aggregation function alternatives; semantic similarity was calculated using symmetrized Resnik, unweighted ancestral overlap, and versions of ancestral overlap weighted by OMIM catalog information content and the topological information specified for the GO-Universal method [39]. Globally, the transitive maximum achieved the lowest median rank. f Comparison of relative performance. Phenotype and filtered genotype data for 47 cohort cases with reported molecular diagnoses were analyzed via the transitive maximum OMIM Explorer algorithms, phenotype-collapsing alternative algorithms, and comparator tools. A minor allele frequency (MAF) filter of 1 % MAF was applied in PhenIX, PHIVE, and hiPHIVE. Because eXtasy limits the quantity of phenotypic inputs to 10, we supplied eXtasy with only up to the 10 phenotypes with the highest information content (i.e., rareness in the OMIM catalog) scores for each case. Via OMIM Explorer, transitive maximum aggregation (Resnik) returned a top ranking for 16/47 = 34.04 % of the cohort and a ranking in the top five for 30/47 = 63.83 %; the overall best alternative, PhenIX, returned a top ranking for 15/47 = 31.91 % and a ranking in the top five for 24/47 = 51.06 %

Back to article page