Use limma for correlation analysis of RNA-seqdata and continous trait

Question

[Yibin]

Hi, there I want to use voom function in limma package to analyze the correlations between expression and age(continous data). I am not sure whether it is suitable to use this package, which is often used for binary data. And the p values were far small compared to pearson correlation test. Please give me some advice. Thank you~ "code"

"voom-plot" "result of top genes"

Answer 1

[Gordon Smyth]

limma and voom are routinely used for continuous covariates like age, but you are not using voom correctly. The voom function is applied to counts rather than to cpm values.

Just use voom as documented, e.g., as in the workflows or in the limma User's Guide, for example:

design <- model.matrix(~age)
keep <- filterByExpr(dge, design)
dge <- dge[keep,,keep.lib.size=FALSE]
dge <- normLibSizes(dge)
v <- voom(dge, design)
fit <- lmFit(v, design)
fit <- eBayes(fit)
topTable(fit, coef="age")

limma is more powerful that computing correlation coefficients and will generally give smaller p-values for the most significant genes.

[Yibin]

That is very kind of your advice. I corrected the mistake in voom() using. but the plot is not smooth , which seems indicate a bad fit of data?

[Gordon Smyth]

There should not be a increasing trend of standard deviation with log2(count size) on the right hand side of the plot. I have never seen a voom plot of RNA-seq data that looks like this so it seems there is something wrong with your data or with the way you have input it to limma. Are you sure that your original data consists of RNA-seq read counts?

[Yibin]

Yes, I used the read counts. I referred to this example before. log2 tranformation was performed in voom. Differential Expression with Limma-Voom

[Picasa]

using a model with a continuous variable age, how do you interpret the logFC ?