Weighted gene co-expression network construction

Given that gene co-expression analysis is extremely sensitive to the existence of abnormal samples, strict quality control procedures were implemented to ensure the highest quality level, followed by step-by-step network construction and module detection. To construct a scale-free gene co-expression network, the WGCNA package in R (14,22,23) was used. First, Pearson's correlation matrices were performed on all gene pairs. Next, using the power function amn=|cmn|β (where amn is the adjacency between genes m and n, and cmn is the Pearson's correlation between genes m and n), a weighted adjacency matrix was constructed. As a soft-thresholding parameter, parameter β may penalize weak correlations between genes while emphasizing strong correlations. To ensure a scale-free network in the present study, the power of β=4 (scale free R²=0.91) was selected (24). Then, the adjacency was transformed into a topological overlap matrix (TOM); TOM is defined as the contiguous sum with all the other genes used for network generation and for measurement of the network connectivity of genes (25). Then, we calculated the corresponding dissimilarity (1-TOM). To classify genes with similar expression profiles into different modules, average linkage hierarchical clustering was performed, according to TOM-based dissimilarity measures; the minimum size (genome) of the gene dendrogram was 50 (26). To investigate the module further, the dissimilarity of module eigengene (MEs) was calculated, a cut line for the module dendrogram was selected, and certain modules were merged (16).