Subtraction of count data

[gilby]

Hi

I have rna sequence data for 4 biological samples: A_stim, A_back, B_stim and B_back. Ultimately, I would like to measure differential expression between 2 experimental conditions A and B. Before I can do this I have to subtract background counts for the two conditions i.e. subtract A_back from A_stim, and B_back from B_stim. I guess before I can subtract, I have to first normalise the samples e.g. convert to RPKM values, and then perform differential expression analysis on the subtracted RPKM values?? Does this seem like a valid way of doing the analysis? I usually use edgeR or DGEseq for differentail analysis, but I don't think they accept RPKM values?

Any thoughts on this would be appreciated.

Thanks,

[mastal]

I am not a statistician, but I would say that what you are trying to calculate
by subtracting normalised background counts A_back from A_stim, is the DE between A_back and A_stim.

Are A and B biological replicates?

[gilby]

Hi mastal,

Thanks for your response.

This may be this is a better explanantion of my problem: let us say I have 2 types of cell, liver and heart, and each type of cell is subjected to 2 treatments, nil stimulation and stimulation. Rna is extracted and sequenced for the 4 samples - for simplicity lets ignore biological and technical replicates for now.
First of all, I would like to subtract the 'nil stimulation' counts from the 'stimulation' counts for each cell type leaving us with 2 sets of counts: liver subtracted and heart subtracted. I would then like to calculate the differential gene expression between the 'heart subtracted' and 'liver subtracted' count data. How would you go about doing this?

Thanks.

[rskr]

I don't think you have to subtract the background, but your model does have to take into account the expectations. A rough example is, if you sequenced twice as many reads in one sample as another you need to calculate the probability that the proportion is significantly different than 2:1. The problem with subtracting the backgrounds, is you have the potential to end up with negative expressions, which doesn't make much sense.