RTqPCR Differential Expression with multivariate linnear models

Question

[Emilio Marmol]

Hello everyone,

So I have some data from RTqPCR experiment in different samples meassuring a number of genes (80-90). I have pre-processed de raw Ct data, correcting for efficiency, normalizing with reference genes with NormFinder and geNorm selected genes, averaging technical replicates and calculating relative quantifications which were finally transformed to log2 scale.

I have one matrix with log2 RTqPCR data for several genes in several samples, each for one different tissue for the same analyzed animals.

The situation is, instead of calculating a simple t-test between two groups for assesing differential expression, I would like to take into account, not only the grouping variable, but also another factorial variable (sex), and a continuous variable, to fit a multivariate regression model for differential expression analysis.

The thing is, I have used limma package for analyzing Microarray data previously, and I was wondering if it would be correct to use the log2 transformed matrices for fitting a limma differential expression analysis, and also how should I define the contrast matrix and formula for linnear regression, if I want to take into consideration the sex (factor) and other continuous variable, apart from the group (two groups).

My phenotype file would be somewhat like:

Samples     Sex    Group    Cont.Var

sample1      1        1        2.3
sample2      1        1        3.4
sample3      1        1        2.5
sample4      2        2        4.6
sample5      2        2        6.2
sample6      2        2        4.1

What formula should I define at design function for making the contrast between group 1 and 2? Say, I would like to make the contrast between groups 1 and 2, taking into consideration the interaction of sex and Cont.Var with the normalized log2 Ct data.

Would it be another approach or recommended pipeline for contrasting this hypothesis?

Many thanks

Answer 1

[Kevin Blighe]

In most cases, Limma just fits a linear model to your data that is ultimately invoking lm.fit(), which is the same function used by lm() (both from Stats package in R). So, all that you really need to do is decide on your model for lm() and then run it independently for each of your genes. With lm(), you can very easily specify any design formula that you want, including with interaction terms.

I give practical examples of this, here:

• Standard regression functions in R enabled for parallel processing over large data-frames (see section 3.2)
• R functions edited for parallel processing

Kevin