Assessing protein-disease association signicance from candidate ranking lists

A. BERENSTEIN; I. IBAÑEZ; A. CHERNOMORETZ

Santiago de Chile

Congreso; 2nd International Congress of ISCB - L.A.; 2012

International Society of Computational Biology

One of the most challenging problems in biomedical research is to understand the mechanisms underlying human diseases. Great effort has been spent on determining genes associated to diseases,from early genetic mapping and molecular biology studies to recent efforts making use of knowledge from human genome sequence projects. All these studies have provided with large data sets on thegenetic basis of human diseases.In this context the network-metaphor has appeared as an appealing framework not only for the sake of data organization, but also to unveil patterns of biological relevance . In recent years there has been a lot of interest in the application of complex network theory in human health related research, in order to analyze molecular basis of human diseases, identify comorbidity patterns, perform drug priorization tasks, and to predict new disease-gene product associations. A large number of this last type of researchprograms assume that protein involved in the same disease have an increased tendency to interact with each other. They involve the use of already known gene-disease associations, a complex network ofinteracting proteins that encodes physical or functional relationships between them, and a kind of information propagation technique in order to rank candidate proteins in terms of their degree ofassociation with the disease-related seeds.In this communication we show that predictions based on ranking candidate lists produced by this type of algorithms can be highly biased by network topology properties and produce inaccurate estimation of protein-disease association significances, and we propose a boot-strapping technique in order to alleviate this problem.