I am a first year PhD student and I am new to RNA-seq analysis. I have some questions about the validity and feasibility of the analysis that my mentor asked me to do. I would appreciate any advice or suggestions from more experienced researchers.
My project involves analyzing RNA-seq data from four samples (n=2) that represent two different conditions (control vs treatment). I used DESeq2 to normalize the data and test for differential expression between the two conditions. I found 27 DE genes between the two conditions, with some relevant results from GO/KEGG enrichment analysis using clusterProfiler.
My mentor was hoping to get more results from the RNA-seq data and wants me to create our own pathways to analyze and estimate changes in cell types. For example, he wants me to assemble a list of genes involved in cholesterol synthesis and plot the log2 fold change for each gene in one large barplot, regardless of significance. Or he wants me to use gene markers of neutrophils to estimate if they are increasing or decreasing between the two conditions.
I am not sure if this approach is valid because it seems to ignore the statistical significance of the DE genes. It also seems to be very subjective and arbitrary, as I would have to choose which genes to include in each pathway or cell type marker set.
If this approach is valid and I am wrong, could someone please explain why. However, if my concerns are valid what resources can you point me to that I can show my mentor. Thanks!
Although your statistical power will be somewhat low with only two replicates, it's may still be ok to run GSEA with caution. Make sure you really dig into the QC steps provided in the edgeR and DESeq2 manuals such as PCA plots to be sure you aren't modeling some sort of technical artifact or introducing problems via an analysis/code error. The cavaet of these follow-up analysis being that since there were so few DEGs GSEA might not turn up anything interesting, and I would be sure to follow-up on findings with orthogonal analysis and benchwork if possible.
Plotting log2FC without regard to p value seems questionable to me. As rpolicastro said, if you back up your findings of these genes with wet lab validation then I'd say that is definitely justified. Maybe qPCR data on control vs treated cells.