Networks and Heart Failure
The technological explosion of methodologies for measuring genome-wide gene expression has resulted in significant advancements for understanding the molecular underpinnings of complex diseases. In particular, efforts for dissecting the complex pathways involved in heart failure (HF), a life-threatening, debilitating condition that costs an estimated $34.8 billion a year, with high prevalence and annual mortality of 10% in the US alone, have been encouraging but insufficient to fully explain this heterogeneous condition. As part of a large effort to obtain a global picture of the transcriptional programs and genetic controllers of HF, we are collecting genotype and cardiac gene expression measurements on a cohort of almost a thousand patients, from ischemic and non-ischemic failing hearts, as well as from donor controls. For the first stage of the project, we utilize array-based genotype and genome-wide gene expression measurements of 316 patients to construct a network view of the transcriptomic programs and pathway structure that drives healthy and HF cardiac states.
We have created novel network-based systems analyses to jointly model gene co-expression relationships and the self-organization of genes into gene "communities": functionally cohesive and potentially overlapping groups of genes. Interpreting network structure using this paradigm is much more intuitive: gene interactions emerge from the self-organization of genes into functional modules. Further, the strength and density of the observed correlations become a function of this community structure, and allow for the interpretation of global network changes between different phenotypic states. Consequently, we use network and community structure for ranking genes , networks, and communities that drive the change from a normal to a failing heart.
This joint inference of community and network structure is in contrast with several methods that already exist for reconstructing the gene networks from expression: most estimate gene modules or communities post-facto, after inferring gene-gene interactions from co-expression. Furthermore, most of these methods yield non-overlapping gene modules, effectively assigning one function for each gene, which is a biological simplification that our methods overcome.