A spectral analysis and network science approach to identify influential diseases based on disease-gene associations

We model a disease-disease network as an undirected graph of diseases (vertices) wherein two vertices are connected with an edge if the corresponding diseases have one or more common associated genes (the number of common genes is the weight of the edge). In such a weighted graph, a disease with a larger number of common genes with several other diseases is more likely to incur a higher eigenvector centrality (EVC). Our hypothesis is that a person with a higher EVC disease is more likely to acquire other related diseases compared to a person with a lower EVC disease. We tested our hypothesis on the disease-disease network constructed from the results of the disease-gene association studies reported in the NIH GWAS catalogue and OMIM database. The disease EVC values exhibited a Pareto distribution (80-20 rule): around 18 % of the diseases have larger and significantly different EVC values and the remaining 82% of the diseases had lower and similar EVC values. This implies that around 18% of the diseases in humans are more likely to have a larger likelihood of leading to other diseases.