Mining human phenome to investigate modularity of complex disorders.

A principal goal for biomedical research is to improve our understanding of factors that control clinical disease phenotypes. Among genetically-determined diseases, identical mutations may exhibit substantial phenotype variance by individual and background strain, suggesting both environmental and genetic mutant allele interactions. Moreover, different diseases can share phenotypic features extensively. To test the hypothesis that phenotypic similarities and differences among diseases and disease subvariants may represent differential activation of correlated feature "disease phenotype modules", we systematically parsed Online Mendelian Inheritance in Man (OMIM) and Syndrome DB databases using the UMLS to construct a disease - clinical phenotypic feature matrix suitable for various clustering algorithms. Using Cardiovascular Syndromes as a model, our results demonstrate a critical role for representing both phenotypic generalization and specificity relationships for the ability to retrieve non-trivial associations among disease entities such as shared protein domains and pathway and ontology functions of associated causal genes.