INTRODUCTION and OBJECTIVES: It has been shown that proteins involved in similar Mendelian and complex phenotypes have a strong tendency to interact directly and physically in human protein interaction networks. In particular, first order interactions have been explored in a number of methods for prioritizing candidates in linkage regions associated with a particular phenotype. However, these strategies lose their power when the loci become too big, perhaps because they are confined to using direct first order interactions or use gene ontology and expression data, to predict higher order physical interactions. General methods for genome-scale prioritization of candidates in a phenotype based on protein interaction network data, have to our knowledge not been reported, particularly in the search for molecular causes of birth defects. In numerous complex disorders, Genome Wide Association studies (GWAS) have incriminated genes in the same disease that are not known to obviously participate in the same cellular pathway. This could be because there is no pathway relationship connecting the genes or because we do not have a complete overview of all biological pathways or knowledge of their crosstalk. If the latter reason is correct, a lack of knowledge on the precise composition of many pathways must be taken into account when constructing models that systematically uses pathway relationships to determine novel components in complex disorders. Here, we present a model that in a given disease, determines if a candidate significantly interacts with known disease causing proteins in higher order interaction networks. A component of this model is refined large-scale proteomics data, meaning it is not confined to or biased towards existing well-known pathways. In this way, our model mirrors the pathway independent discovery in genome-wide association studies. This model has the power to make accurate genome-wide predictions of risk factors in a complex phenotype.
Showing the most recent 10 out of 20 publications
