I am working to perform a TRAP-seq analysis, and I am struggling with a decision on what is the best method for analysis. Assuming a standard TRAP-seq analysis as described in the following table:
assay treatment ind ind_n sample1_IP | IP | control | 1 | 1 sample1_input | input | control | 1 | 1 sample2_IP | IP | control | 2 | 2 sample2_input | input | control | 2 | 2 sample3_IP | IP | control | 3 | 3 sample3_input | input | control | 3 | 3 sample4_IP | IP | treated | 4 | 1 sample4_input | input | treated | 4 | 1 sample5_IP | IP | treated | 5 | 2 sample5_input | input | treated | 5 | 2 sample6_IP | IP | treated | 6 | 3 sample6_input | input | treated | 6 | 3
Without having read any of the literature on the analysis, I thought that I could look for the differences in IP vs. input on the effect of treatment. However, since in TRAP-seq the IP and input are paired by sample, I thought that we should use a paired design, such as:
~ assay + treatment + treatment:ind_n + treatment:assay. Upon talking with the researcher, they were expecting a t-test, which didn't seem the best method to me, but my statistical background is quite limited, so I started looking through the literature. I feel like I have found just about everything, including t-test. I see support for the use of
assay + treatment + treatment:assay from this protocols paper, which is relatively recent and seems reasonable with good documentation. However, I am uncertain why there is no inclusion of the paired design.
My main question is whether DESeq2 is still recommended and accurate for determining translational efficiency? And if so, should I be correcting for the paired nature of the samples. Are there other considerations in this specific type of data that would make this not the best method of analysis. I am sorry if there is something key that I am missing.
Thank you for your time and help,