For about 30% of hereditary diseases no disease gene is currently known. Very little if anything at all is known about the molecular basis of these orphan diseases. In this Thesis I present an ontological method that accurately quantifies similarity between heritable diseases modules in the interactome, which can be used to help pinpoint the location of the perturbation causing the orphan diseases . This method, based on the MeSH ontologies, effectively brings together the existing information about diseases that is scattered across the vast corpus of biomedical literature. I prove that sets of MeSH terms provide a highly descriptive representation of heritable disease and that the structure of MeSH provides a natural way of combining individual MeSH vocabularies. I also show that the measure can be used effectively in the prediction of candidate disease genes. The effective use of the vast information available allows the measure to be applicable for orphan diseases: the measure can help pinpoint the location of their molecular perturbations. More generally, the measure enables the transfer of knowledge between similar diseases, providing hypotheses for disease genes and even suggestions for drug repositioning. I have validated the method through a machine learning approach to show the predictive power of the measure. Further to the numerical evaluation, I have curated a highly illustrative set of examples for the literature showcasing the accuracy of the method. Lastly, I show that the measure is effective for the prediction of candidate disease genes. I have developed a web application to query more than 28.5 million relationships between 7,574 hereditary diseases (96% of OMIM) based on the similarity measure. During my PhD I have also developed GOssTo and GOssToWeb a console and web application to compute semantic similarities in the Gene Ontology. GOssTo was integrated into a disease gene prediction pipeline that showed the advantages of using functional similarities to improve the predictions.
展开▼
