csuWGCNA: a combination of signed and unsigned WGCNA to capture negative correlations

Abstract

Network analysis helps us to understand how genes jointly affect biological functions. Weighted Gene Co-expression Network Analysis (WGCNA) is a frequently used method to build gene co-expression networks. WGCNA may be calculated with signed or unsigned correlations, with both methods having strengths and weaknesses, but both methods fail to capture weak and moderate negative correlations, which may be important in gene regulation. Combining the advantages and removing the disadvantages of both methods in one analysis would be desirable. In this study, we present a combination of signed and unsigned WGCNA (csuWGCNA), which combines the signed and unsigned methods and improves the detection of negative correlations. We applied csuWGCNA in 14 simulated datasets, six ground truth datasets and two large human brain datasets. Multiple metrics were used to evaluate csuWGCNA at gene pair and gene module levels. We found that csuWGCNA provides robust module detection and captures more negative correlations than the other methods, and is especially useful for non-coding RNA such as microRNA (miRNA) and long non-coding RNA (lncRNA). csuWGCNA enables detection of more informative modules with biological functions than signed or unsigned WGCNA, which enables discovery of novel gene regulation and helps interpretations in systems biology.