RNA seq and Invitro data not matching

Question

[Bioinformatician_in_trouble]

Hi all, Invitro data shows upregulation of certain genes where as RNA seq analysis shows a positive foldchange for same genes but not significant p value and q value. How can i justify this?

[ATpoint]

What is Invitro? qPCR? Please add details: How many replicates, what is the setup, does PCA indicate good reproducability between replicates? What is the tool you used for analysis? Also please show an MA-plot and highlight the gene you are interested in. Your experiment could simply be underpowered.

[Mark]

More info needed. How was RNAseq performed and analysed, what's invitro data and how was it analysed.

Answer 1

[i.sudbery]

All experimental technique have a limit to their precision. Thus, the estimated foldchange we get out of both qPCR and RNAseq is just that - an estimate. That is why we put error bars on things - the error bars (at least if they represent the 95% confidence interval, which they normally should) more or less tell us the range of values that could plausibly be true. In RNAseq, these error-bars can be really quite wide if there is a small number of replicates or for a lowly expressed gene. The fact that this gene is not significant in RNA-seq analysis means that the error-bars cross 0. That is both negative and positive logfold changes could plausibly be true. This is what it means for a gene to not be significant. If you did yuor analysis in DESeq2, then the standard error is in the output table and you can calcaulte the 95% confidence interval for the fold change. Here is an extract from the DESeq2 vignette:

    ##                     baseMean      log2FoldChange             lfcSE
    ##                    <numeric>           <numeric>         <numeric>
    ## FBgn0000008 95.1442917575889 0.0227644122547027  0.22372865161848
    ## FBgn0000014 1.05652281859341  -0.495120386253493  2.14318579304427
    ## FBgn0000017 4352.55356876647   -0.23991894353759 0.126336905404352

For FBgn0000008 the best guess log2FoldChange is 0.022, but the standard error is 0.223. The 95% CI is very approximately 2SE, so the range of log2FoldChanges that could feasibly be true is 0.022 -/+ 20.22 = 0.022 +/- 0.44 or -0.42 to +0.46. Of course, all this also applies to results from qPCR.

[i.sudbery]

I changed this answer because I miss-read the question and thought the "invitro" data and the RNAseq were going in opposite directions.