I am a PhD student who inherited some log2cpm data of expression data from bulk kidney tissue from a UUO(unilateral urethral obstruction) experiment that tests a new drug. The sample material consists of:

  • 6 x Ligated Kidney (Untreated
  • 6 x Ligated Kideny (Treated)
  • 3 x Unligated Kidney (Untreated)
  • 3 x Unligated Kidney (Treated)

The previous study showed that this drug was effective against fribrosis, so the aim of my study is to investigate how this drug affects inflammation and mitochondrial function in the diseased kidney.

So I have filtered my original dataset based on MitoCarta v2 DB, so investigate how the drug affected mitochondrial function. I afterwards run a PCA analysis on log2cpm data which revealed the following: enter image description here

The following figure shows PC1(84.51%) and PC2(4.69%). Red (Ligated, Untreated), purple (Ligated, Treated), green(healthy kidney, not treated), orange (Healthy kidney, treated).

I interpret this as there is a very small difference between treated and not treated ligated kidneys, however there is a small difference.

I have just started doing bioinformatics and I am very unsure which approach I should choose further, however I have though maybe do a highly variable genes analysis (HVG) see link, isolate the 500 most variable genes and run a heatmap with hieraichal clustering. I was also thinking of looking at the loadings in PC2, and take this further. The truth is that I am very unsure what is the right approach, and my research group does not have bioinformatician I can ask anymore.

I have the following questions:

  • What kind of analysis you guys would suggest I use to investigate further (Hieraichal clustering, HVG etc?)
  • Is there any alternative preprocessing I could use before I run the PCA, to make the analysis more accurate?
  • Are there any packages that you can suggest?
  • Any other tips?

Any other tips?

I asked this same question on Biostars as well https://www.biostars.org/p/367722/#367730

  • $\begingroup$ Welcome to the site. What is your expected result? A list of genes to then validate experimentally with an immuno* or some list of genes for prediction of status/response... What other information do you know about those samples: Could it be that some samples are before and after treatment? Have you considered analysing the differentially expressed genes on the whole dataset (without filtering first for mitochondrial related genes)? $\endgroup$
    – llrs
    Mar 6, 2019 at 8:49
  • $\begingroup$ It looks like you'd be best served with a standard differential expression analysis using a factorial design. You should get the raw counts data for that though. $\endgroup$
    – Devon Ryan
    Mar 6, 2019 at 9:23
  • $\begingroup$ Hey, the Animals were sacrificed on day 7 or day 15 after the surgery, and treatment started immediatley after surgery. Drug was given immediatley after surgery. The reason I want to investigate mitochondrial and immune genes, is that I previously did a GSEA with TopGO package of the whole transcriptome based on logfc-values and q-values (down and upregulated genes isolated, qval-cutoff 0.01, (+/-)fc 1.5) of Ligated(untreated vs. healthy kidneys), and Ligated(treated) vs. Ligated(untreated) kidneys. see figure. It showed that mosltey mitochondrial and inflammatin was affected $\endgroup$ Mar 6, 2019 at 9:37
  • $\begingroup$ Now I want to investigate a bit further what molecular pathways the drug affects and what constitute the improvement. $\endgroup$ Mar 6, 2019 at 9:38
  • $\begingroup$ @NewbieCoder Did you include the time in the design matrix when analysing differential expressed genes? Also if you are interested in those pathways is different than being interested in the genes expressed in the mithocondria as it appeared first. Could you edit the question and clarify what is your question? $\endgroup$
    – llrs
    Mar 6, 2019 at 16:19

1 Answer 1


Since you indicate already having performed differential expression and using gene set enrichment the next step is to look at what changes due to drug treatment in the pathways of interest while having a decent understanding of those pathways. Things like IPA or pathview (from Bioconductor) can aid in this, but at the end of the day you need to have a decent understanding of the pathways yourself to judge what seems to be like the best targets for future work.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.