CWGCNA: an R package to perform causal inference from the WGCNA framework

Abstract

WGCNA (weighted gene co-expression network analysis) is a very useful tool for identifying co-expressed gene modules and detecting their correlations to phenotypic traits. Here, we explored more possibilities about it and developed the R package CWGCNA (causal WGCNA), which works from the traditional WGCNA pipeline but mines more information. It couples a mediation model with WGCNA, so the causal relationships among WGCNA modules, module features, and phenotypes can be found, demonstrating whether the module change causes the phenotype change or vice versa. After that, when annotating the module gene set functions, it uses a novel network-based method, considering the modules' topological structures and capturing their influence on the gene set functions. In addition to conducting these biological explorations, CWGCNA also contains a machine learning section to perform clustering and classification on multi-omics data, given the increasing popularity of this data type. Some basic functions, such as differential feature identification, are also available in our package. Its effectiveness is proved by the performance on three single or multi-omics datasets, showing better performance than existing methods. CWGCNA is available at: https://github.com/yuabrahamliu/CWGCNA.