Differential expression analysis of few genes from rnaseq or microarray

Question

[ASid]

So,there is a project I am working on with somebody for analyzing deferentially expressed genes from couple of array/RNA seq data sets. Seems like people are interested in only checking up the expression profile of only two genes among different comparison groups in all of the data sets. Just to be clear i used a straightforward limma based deferentially expressed gene analysis.

So at the end we have these fold change values along with the t , p and adjusted p values (for multiple comparisons) for each datasets and the comparison that we performed.

Now we take out just these statistics for these two genes and use them for further analysis.

According to my understanding we should use adjusted p values for any of the further analysis because we used limma and it already used moderated t test where the variances of all genes are considered in order to generate the final t p and adj p values. plus the fold change is a intensity based relative measure and we can not kind of compare this with RT-pcr or simple pcr based analysis type of thing. what i mean to say is people want to use p values and i am not sure about it. Their argument is that when we use RTpcr or pcr based low throughput techniques we usually end up analyzing just bunch of molecules and we dont use ofcourse adj p values and since we are interested in only two genes from array analysis so we should use p values instead of adjusted p values?

Any possible explanation or opinion on this (positive, negative, anything) ?

Thank you.

[dsull]

Here are some people's thoughts (which I generally agree with): http://seqanswers.com/forums/showthread.php?t=48011

I quote: "You cannot simply adjust the p-values from the original analysis as those p-values were based on a variance model from the whole transcriptome data set. Your small subset of genes may represent a very different set of data than the original complete data set."

Here are my own thoughts from a previous post I made on biostars (if you're considering using a standard non-limma-based Student's t-test on your microarray data): C: p-value in Limma vs Graphpad

In essence, as has been stated by other comments, use adjusted p-values if you're going off your limma analysis. Also, think about what do you really want to gain from looking at a (adjusted) p-value. People think that p-values are the "end all be all" (they're not; all they do is tell you if there's enough evidence against the null hypothesis).

[h.mon]

Now we take out just these statistics for these two genes and use them for further analysis.

It is very unclear what you mean here. Can you clarify how t-statistics, p and adjusted p values are used for "further analysis"?

[ASid]

Yeah so further analysis is nothing more than that these statistics are coherent with the hypothesis they were testing coming up as significant for related metabolites modulated by related genes(coming from microarray) of similar comparisons performed separately and we are gona see them in plots!

Answer 1

[jared.andrews07]

So is your question whether to use adjusted p-values or not? It's not really clear what your "further analysis" is/would be. In general, yes, you should use the adjusted p-values from analyses where multiple-test correction is applied (like microarrays), regardless of whether you're interested in only a handful of genes or not. This sounds like a situation where folks want you to use the standard p-values because they're generally lower and thus may let them say what they want with the protection of it being "significant".

Don't fall into such a trap, as it may leave you chasing ghosts experimentally or draw the ire of observant reviewers.

[ASid]

yes my question is exactly this!whether to use adjusted p-values or not? I already agree with you on this and want to know that my above observation and explanation is correct regarding limma taking into account all gene variances to come up with statistical values and thus picking the the p values of few genes by hand can let them fall into the false positive category?correct ? Plus microarrays are relative intensity based and thus cant be compared like any other low througput assay measuring value ?right?

I need convincing statistical explanation for this to persuade them! May be I am not good at explaining but want to know this is correct explanation?

[jared.andrews07]

This paper explains multiple testing and why it's important very clearly with simple examples. It should provide a strong foundation for your argument and give you some ideas for how to explain it to less statistically-minded people in a way they can understand.