0
$\begingroup$

I'm trying to compare the gene expression data reported in the studies documented in GEO for a specific gene expressed in a tissue of my interest in Homo sapiens. I compared the values reported in 4 studies, in two studies the expression value of gene A lies between 15-30, in the other two studies the range lies between 1-4.

I would like to ask for advice on whether it is meaningful to compare the data from different studies. What are the factors that one should keep in mind while making such comparisons?

Any help would be highly appreciated

Improve this question
$\endgroup$
4
  • $\begingroup$ From the suggestions below, as a first check, I have looked for the platform used for the microarray analysis. The platform varies in each study. I'm trying to compare the probe densities. Unfortunately, I am not able to find this information.Any suggestions? $\endgroup$
    – Natasha
    Commented Aug 16, 2018 at 13:04
  • $\begingroup$ You won't be able to make comparison in this case. $\endgroup$
    – SmallChess
    Commented Aug 16, 2018 at 13:39
  • $\begingroup$ Could you please explain why? $\endgroup$
    – Natasha
    Commented Aug 16, 2018 at 16:23
  • $\begingroup$ The two answers here. $\endgroup$
    – SmallChess
    Commented Aug 16, 2018 at 20:54

2 Answers 2

Reset to default
1
$\begingroup$

There's going to be a batch effect between studies, so unless you have some sort of constant group that you can use for between-study normalization you're going to completely muck up your results from using data from multiple studies.

More specifically, this is what packages like sva and the combat() command in it are used to correct. Note that you should go back to the raw .CEL or .fastq files if you want to process multiple studies, in order to at least ensure that everything is processed in the same way.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Please check my recent comment in response to your answer $\endgroup$
    – Natasha
    Commented Aug 16, 2018 at 13:05
1
$\begingroup$

It doesn't make sense to compare gene expression data from different studies unless you have a negative spiked-in control such as sequins.

The most obvious thing you will need to do is normlizate the data for various bias such as batch effects, sequencing platforms, sequencing depth, bioinformatic workflows and more ... Your task will be quite complicated as you must justify how your normalization would work for data from different studies.

You should check how the raw data was generated (methods section), convince yourself, and then apply gene quantification yourself from the raw files.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Please check my recent comment in response to your answer $\endgroup$
    – Natasha
    Commented Aug 16, 2018 at 13:05

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.