Weighted gene co-expression network meta-analysis

For the SA Branch’s July meeting Max Moldovan from the South Australian Health and Medical Research Institute (SAHMRI) spoke about the application of weighted gene co-expression network analysis (WGCNA), a common tool for the detection of co-expressed gene groups.  Max proposed meta-analysis of RNA-seq datasets within WGCNA as a way of overcoming problems associated with insufficient sample sizes of RNA-seq datasets.  He illustrated numerically the gain in co-expression estimates precision from combining several datasets within the WGCNA framework.

Since 2014, Max Moldovan has worked in the South Australian Health and Medical Research Institute (SAHMRI) being involved in Infection & Immunity and prostate cancer research. RNA-sequencing (RNA-seq) is a technology allowing analysis of gene expression levels assessed throughout an entire genome of a biological organism. At the gene network level, it is regularly required to assess dynamics of gene expression not only for individual genes, but also for groups of genes suspected to systematically change expression in a collective way, or being co-expressed. Weighted gene co-expression network analysis (WGCNA, Langfelder and Horvath, 2008; BMC Bioinformatics 9:559) is a common tool for the detection of co-expressed gene groups. In this talk, Max proposed meta-analysis of RNA-seq datasets within WGCNA as a way of overcoming problems associated with insufficient sample sizes of RNA-seq datasets. Max illustrated numerically the gain in co-expression estimates precision from combining several datasets within the WGCNA framework.

As a numerical example, Max’s simulation study generated samples of K = 5, 10, 30 correlated genes with known correlation, 5000 MC replications. Sample sizes for each replication were randomly sampled and the resulting simulated datasets were meta-analysed using alternative strategies: “just-merge”, “batch-and-merge” (apply a batch adjustment before merger of individual datasets) and “metaphor” (a random-effect model meta-analysis of correlation coefficients implemented through the matafor R package).

The WGCNA meta-analysis of several RNA-seq datasets appeared to reveal interesting co-expression patterns, as was illustrated with empirical datasets from prostate cancer tissue. For the simulation the batch-and-merge meta-analysis strategy led to more precise estimates, and for the empirical datasets this strategy also produced richer biological annotation relevant co-expression modules.

Max proposed further analysis involving a comprehensive study of the resulting network (e.g. analys

Max Moldovan

is of hub genes, bottlenecks, centrality etc. using R igraph package).

By Paul Sutcliffe

 

 

 

Follow

Get the latest posts delivered to your mailbox:

Show Buttons
Hide Buttons