The field of network medicine represents the marriage of systems biology with network science to try to understand the causes of human disease. Channing Division of Network Medicine (CDNM) network analysis investigators embrace this approach to human health by using network analysis techniques to holistically evaluate complex biological systems. This stands in contrast to the “reductionistic” approach often seen in medical research.
Collectively, our group has applied a broad range of network analysis approaches to study multiple diseases and biological systems. Reflecting the history and composition of CDNM cohorts, much of this work has been in COPD and asthma.
Network medicine embraces the fact that human phenotypes and pathophenotypes (or disease phenotypes) are driven by complex interactions among a variety of molecular mediators. In other words, disease processes aren't driven by single elements or single entities but by complex interacting systems.
Characterizing these systems using the approaches and mathematical tools provided by network science can often help us to better understand how to control and treat disease.
Protein interaction networks are built based on knowledge of physical or functional interactions between biological molecules. The connectivity pattern among proteins in this network can provide key insights into biological systems and allows us to hypothesize how perturbations to these protein interactions may impact disease. For example, disease-associated genes (such as those implicated in genome-wide association studies [GWAS]) often are localized within a specific neighborhood of the molecular interaction network, forming a “disease module” and implicating other nearby genes in the disease.
Formally, a Bayesian network is a directed acyclic graph whose nodes represent probabilistic random variables. Bayesian networks encode conditional dependencies between a variable and its parent nodes such that the variable is independent of its non-descendants given its parents.
Bayesian networks are attractive models since their edges can encode the influence of undetected variables, which helps to overcome the imperfect knowledge and incompleteness of experimental data in biological systems.
