How to best represent three biological replicates of single tissue sample?

Question

[BioinfoBee]

Hello All,

Curious as to how to best represent three biological replicates of a single tissue sample of RNASeq data? Assuming that these each replicates are normalized (RPKM, TPM), is there any other statistical analysis that needs to be taken into account before deciding on taking average (mean), median or other approaches?

Regards,
B

Answer 1

[Asaf]

What you have is called clustered data. There are methods to deal with the potential bias it has on estimating the variation like using mixed models and having the tissue a random effect on the intercept. Tools like DESeq2 or edgeR don't include random effects, however, and the clusters can be incorporated as fixed effects using edgeR trick for nested effects.

And applying the statistical test on the raw counts is preferable (with the right model).

[BioinfoBee]

Thank you Asaf for your suggestions. Still not clear on if it's allowed to take an average (mean) of three biological replicates to represent single sample?

For example,

        S1.1     S1.2       S1.3
Gene1   0        0          9
Gene2   3        3          3
Gene3   2        6          1

In the above case, the average(mean) for all three gene is 3. However, I assume that Gene1 is not expressed since only one replicate has read counts which may be due to technical error,

[Asaf]

If anything you should sum the counts. You can do whatever you like but agglomerating the replicates depict the purpose of having biological replicates which is to estimate the variance. Every statistical test relies on the variance (in one way or another), it's usually estimated so as you have more replicates the estimation is better.

[BioinfoBee]

Asaf Thank you again. So for three biological replicates what statistical test do you think suits better?

[Asaf]

If you have the count matrix I would use DESeq2 or edgeR