Hi,
I am analysing RNA-seq data of two genotypes (A and B) in order to detect the transcripts that are involved in a physiologic change that takes place earlier in genotype A and later in genotype B. For genotype A there are 2 time points (A0 and A1) and for genotype B there are 3 time points (B0, B1 and B2), with 3 replicates each (15 samples in total). The reason for not having A on time 2 is that by time 1 such physiologic change has already been produced. (The change has already been produced on A1 and on B2.)
I performed abundance estimation using RSEM and I want to perform differential expression analysis using limma. I am not sure this can be considered a time course experiment as I only have 2 samples for genotype A and I don't have the same number of time points for both genotypes. My idea is to select the transcripts that are differentially expressed between A and B on time 1 (A1-B1) and the transcripts that are not differentially expressed between A1 and B2. However, selecting the transcripts which are significantly not differentially expressed is complicated, since transcripts not appearing in the list of differentially expressed are not necessarily not-DE.
Is there a way that to write my contrast matrix in order to find the transcripts I am interested in?
Thank you for your attention.
Thank you for your answer. I should probably have mentioned that I have used limma for factorial designs before for other experiments. I am able to find the differentially expressed transcripts between conditions. But in my case I am interested in the intersection of transcripts which are DE in (A1-B1) and NOT DE in (A1-B2). I am also aware of your answer about using confidence intervals to a user trying to find similar expressed genes (extract similar expressed genes (SEGs) rather than DEGs ) and I will probably do so, but I wanted to know if there was a simplest way to do this.
There's no formal way to select not-DE genes. In practice, one has to choose genes-that-are-not-even-close-to-significant to fill this role.
Thanks, I wanted to be sure there is no other way to do it. I understand I need to select genes with a wide confidence interval and a small logFC to find non-DE genes between A1 and B2.
No, you don't need genes with a wide confidence interval. What you need is genes with a large p-value. The ideal non-DE gene would have a large p-value and a narrow confidence interval.