A similarity-based method for genome-wide prediction of disease-relevant human genes.

A method for prediction of disease relevant human genes from the phenotypic appearance of a query disease is presented. Diseases of known genetic origin are clustered according to their phenotypic similarity. Each cluster entry consists of a disease and its underlying disease gene. Potential disease genes from the human genome are scored by their functional similarity to known disease genes in these clusters, which are phenotypically similar to the query disease. For assessment of the approach, a leave-one-out cross-validation of 878 diseases from the OMIM database, using 10672 candidate genes from the human genome, is performed. Depending on the applied parameters, in roughly one-third of cases the true solution is contained within the top scoring 3% of predictions and in two-third of cases the true solution is contained within the top scoring 15% of predictions. The prediction results can either be used to identify target genes, when searching for a mutation in monogenic diseases or for selection of loci in genotyping experiments in genetically complex diseases.