gene expression x phenotype correlation

Question

[demoraesdiogo2017]

Hello

I would like to find biomarkers that correlate with geometric mean of cell size and gene expression. I have not found any article that did something similar yet. I tried Spearman's correlation test with TMM normalized expression and one of the most correlated genes was, indeed, a canonical regulator of this phenotype. I assume other significantly correlated genes could also be biomarkers of the phenotype. Because my data comes from humans and has heterogeneity, I intend to stratify this analysis by other phenotypes I am not interested in, and choose genes that are correlated in all, or at least most analyses. To reduce the number of genes to be analyzed I performed the analysis only in the 5000 genes with most variance and with expression of at least 25TPM in 50% of the samples. Does that sound adequate?

Also, is there any reason I should use linear regression or WGCNA instead? I am assuming an univariate linear regression will yield similar results compared with Spearman correlation. From what I have read about WGCNA and as a colleague showed me, the continuous variables are turned into traits, which I believe can be a problem since the cell size range is pretty wide.

Answer 1

[Gordon Smyth]

Since you've used edgeR for TMM normalization already, why not continue with edgeR to correlate expression with cell size? Correlating expression with phenotype (either discrete or continuous) is what edgeR is designed to do. See for example the case study in Chapter 4.8 of the edgeR User's Guide that analyses a drosophila time course experiment. The same methods could be used for cell size as are used in the case study for time.