Seminar
Jun Dong
A Geometric Interpretation of Gene Co-Expression Network Analysis
Gene co-expression network analysis has been used to address systems biologic questions in lower organisms,
evolutionary studies, cancer genetics, and in complex disease gene mapping. How gene co-expression network
analysis relates to more traditional statistical analysis techniques remains an outstanding theoretical question.
By decomposing the gene expression data of a module with the singular value decomposition, we provide a
geometric interpretation of gene co-expression network analysis. These insights have important theoretical and
practical applications. We describe when the network adjacency (pairwise connection strength) between module
genes can be factored into gene specific contributions, referred to as eigengene conformity. The eigengene
conformity of a gene is determined by the absolute value of the correlation between its expression profile and the
module eigengene. We provide an intuitive geometric interpretation of network concepts such as intramodular
connectivity (degree), heterogeneity, density, etc. We show that the module eigengene can be interpreted as the
most highly connected intramodular hub gene. We show that a microarray sample trait (e.g. survival time) gives
rise to highly related measures of module-, hub gene-, and eigengene- significance. We illustrate our results
using three microarray data applications (human, mouse, and yeast). The geometric interpretation facilitates the
derivation of several theoretical results about intramodular hub genes. These theoretical results have important
practical uses including a fuzzy module annotation method.
