RNA-seq design advices

Question

[fmazzio1]

Hi all!

I am looking for suggestions on how to analyze my bulk-RNAseq data. Briefly, I have 87 samples of sorted CD4 and CD8 T cells, two timepoints (pre-chemotherapy and post-chemotherapy), two sites (bone marrow (BM) and peripheral blood (PB)), three groups (non-responders (NR) to therapy, responders (CR) to therapy, healthy donors(HD)). So for patient 001 (assume he/she is a complete responder) I'll have: 001_CD8_pre_BM_CR, 001_CD8_post_BM_CR, 001_CD4_pre_BM_CR, 001_CD4_post_BM_CR, 001_CD8_pre_PB_CR, 001_CD8_post_PB_CR, 001_CD4_pre_PB_CR, 001_CD4_post_PB_CR (8 samples per patient, 4 samples per HD since they have only one timepoint).

The main question we want to answer is what are the DEG across the groups (NR, CR, HD), then if this difference is present pre- and/or after chemo.

My approach would be to split my dataset in 8 subsets (CD8_pre_BM, CD8_post_BM, CD4_pre_BM, CD4_post_BM, CD8_pre_PB, CD8_post_PB, CD4_pre_PB, CD4_post_PB) and then use a Wald to test CR vs NR vs HD in each of them. Another approach could be to do not split according to the timepoint and then make a design as ~ group + timepoint.

Do you think it's appropriate? Other ideas?

Thank you very much

Francesco

[rpolicastro]

You have 4 factor levels: cell, timepoint, site, and group. Right now it's unclear what exact comparisons you want to make. Mentioning all factor levels, can you give examples of some desired comparisons?

[fmazzio1]

For the two cell types (CD4 and CD8) I would make:

CD8 group comparison: CR vs NR vs HD; group, timepoint comparison: CR_pre VS NR_pre VS HD, CR_post VS NR post VS HD; group, timepoint, site comparison: BM_CR_pre VS BM_NR_pre VS BM_HD, BM_CR_post VS BM_NR post VS BM_HD, PB_CR_pre VS PB_NR_pre VS PB_HD, PB_CR_post VS PB_NR post VS PB_HD;

Same thing for CD4

Additionally, since I have a multilevel factor (group = CR, NR, HD), would you use a LRT test instead of Wald?

Thank you!

Francesco

Answer 1

[swbarnes2]

My approach would be to split my dataset in 8 subsets

In general, you don't want to do that. Well, maybe split up the tissues to process separately, but that's it. You want all the samples together to get better dispersion estimates. You can compare one subset of samples to another subset of samples without chopping up the dataset.

[fmazzio1]

Thank you! I'll follow your suggestion!