Combining protein-protein interaction (PPI) network and sequence attributes for predicting hypertension related proteins

Abstract

Cardiovascular disease is set to become the number one cause of deaths worldwide. It is therefore important to understand the etiologic mechanisms for hypertension, in order to identify new routes to improved treatment. Human hypertension arises from a combination of genetic factors and lifestyle influences. Here we study hypertension related proteins from the perspective of protein-protein interaction (PPI) networks, pathways, Gene Ontology (GO) categories and sequence properties. We find that hypertension related proteins are not generally associated with network hubs and do not exhibit high clustering coefficients. Despite this, they tend to be closer and better connected to other hypertension proteins on the interaction network than we would expect, with 23% directly interacting. We find that molecular function category 'oxidoreductase' and biological process categories 'response to stimulus' and 'electron transport' are overrepresented. We also find that functional similarity does not correlate strongly with PPI distance separating hypertension related protein pairs and known hypertension related proteins are spread across 36 KEGG pathways. Finally, weighted Bagged PART classifiers were used to build predictive models that combined amino acid sequence with PPI network and GO properties. © Springer-Verlag Berlin Heidelberg 2008.