5

I have a expression data from a small cohort of samples taken at baseline and after 2 independent treatments. I can do differential expression contrasting T1 and T2 or I can contrast T1 vs baseline and T2 vs baseline and look at the differences. What is the difference between these two analyses? Are they both valid? If so what inferences can be drawn from each?

Nitro
  • 153
  • 4

2 Answers2

4

If you're interested in looking at the differences between two treatments then you'll end up wanting to do both a direct contrast as well as the individual comparisons to baseline.

The direct contrast will give you the genes actually differentially expressed between the two conditions. In practice, you may want to filter this a bit so you only have genes differentially expressed vs. baseline in at least one condition (e.g., to get rid of genes only slightly higher due to T1 and slightly lower in T2, but not different enough in either case to be DE). Use a pretty lax p-value threshold (e.g., 0.1 or 0.2) for this.

Devon Ryan
  • 19,602
  • 2
  • 29
  • 60
  • Thanks. What is your opinion on genes that may be significant in just 1 of the comparisons to baseline but not significant in the direct T1 vs T2 contrast? – Nitro Oct 30 '17 at 20:05
  • 1
    It there's no difference in the T1 vs. T2 contrast then there's no difference. It doesn't matter if you happen to get a significant difference in one of the conditions vs. baseline. You only want to use the comparisons to baseline for filtering your T1 vs. T2 results (assuming you have enough DE genes to bother doing so). – Devon Ryan Oct 30 '17 at 20:22
  • One more question. So if I wanted to do a pathway analysis and compare the two treatments that way. When I generate my gene lists how should I treat genes that are DE in only one treatment vs baseline but not DE in the direct T1 vs T2 contrast? – Nitro Oct 30 '17 at 21:21
  • Treat them as not DE. – Devon Ryan Oct 30 '17 at 21:24
  • So any genes that have similar responses from baseline in both treatments get discarded? Wouldn't you be missing a lot of context if a gene has to be DE vs baseline in both treatments and also DE directly with each other to make either final list? – Nitro Oct 30 '17 at 21:31
  • Yes, any genes with similar responses should get discarded, since they're not different. A gene doesn't have to be DE in either treatment vs. baseline to be DE between them. That's also why I suggested using a very lax p-value threshold if you want to do any filtering of the T2 vs. T1 results. – Devon Ryan Oct 30 '17 at 21:37
  • Sorry, I meant doing a pathway analysis of each treatment vs baseline separately to find the raw response of each treatment. Since they are different from baseline I would still include them in only their respective pathway analysis? – Nitro Oct 30 '17 at 22:02
  • 1
    Yes, certainly include them in such cases. – Devon Ryan Oct 30 '17 at 22:06
  • Cool, thanks a lot for your help. I really appreciate it. – Nitro Oct 30 '17 at 22:09
3

Depending upon the type of treatment used the set of DEGs will change.If the treatments have similar kind of effect you will get a small list (less variable genes will have higher p-val) using a cutoff of p<0.05. So, it's better to start with control vs treated comparisons then T1 vs T2.

Compare the lists form CTRL vs T1 and CTRL vs T2 you will get the genes that are expressed in both conditions as well as unique to individual treatments.

arup
  • 604
  • 5
  • 15
  • 1
    If a gene has a p-value of 0.049 in CTRL vs. T1 and 0.05 in CTRL vs. T2 you would be categorizing it as uniquely DE in one condition, when there's no actual difference between the conditions. – Devon Ryan Oct 31 '17 at 07:12
  • Yes, it's a tricky situation. – arup Oct 31 '17 at 13:45
  • I agree it is a trick situation in my case it is modes of exercise so they are extremely similar responses. The T1 vs T2 comparison yields almost nothing where the vs baseline comparisons are quite different. – Nitro Oct 31 '17 at 16:49