Mining the diseasome

Diseases are traditionally considered as discrete entities and classified accordingly. However, the networks of genes accountable for particular disease phenotypes most certainly overlap, with individual genes simultaneously serving the cause of multiple disorders [5,7]. Clinically distinct diseases have genes in common, like nodes in a network have links in common, and DNs capture this analogy by representing diseases with nodes and the genes they share with links. In such a network representation, breast cancer and pancreatic cancer for instance are two nodes connected by TP53 [5]. What the concept of DN implies is that many susceptibility loci hitherto associated to distinct diseases are in fact likely to contribute to the genetic architecture of several disorders. Hence, rather than initiating genetic association studies with no a priori hypothesis about where in the genome to look for potential candidate risk loci, the information captured by HDNs may serve the purpose of anchoring the search for susceptibility loci in genomic regions known to harbor genetic variants predictive of other "linked" diseases. Subsequently, the human interactome [6], i.e., the compendium of molecular, phenotypic and genetic interactions, or genome-wide regulatory networks [8] can serve as maps to navigate the genome in search of further susceptibility loci.Additional indices on where to start exploring the genome for susceptibility loci can be inferred from general principles of human diseases and clinical data. For example, a considerable fraction of diseases with onset early in life appear to result from defects in enzyme-encoding genes, whereas diseases with onset during adulthood appear to be caused by alterations in genes encoding modifiers of protein functions [9]. Thus, clinical information such as age at onset or severity can serve as valuable expert knowledge to narrow down the genomic search space to genes or genetic domains that are biologically and clinically meaningful.