Is it possible to combine micro array data and RNA-Seq data and perform meta analysis for differential gene expression analysis.

Question

[rajasekargutha]

I am performing meta-analysis for identifying differential gene expression of microarray data from similar experiments. I want to include some RNASeq data to the above data for identifying differential gene expression by the Rankprprod approach. Is it possible to combine micro array data and RNA-Seq data and perform meta analysis for differential gene expression analysis.

Answer 1

[oneillkza]

This is a bit old, but it came up as one of the top Google results for how to combine microarray and RNA-seq data in a meta-analysis.

Anyway, in answer to the original question, this paper from August 2017 did exactly that: https://academic.oup.com/nar/article/45/17/9860/4084660

Note that they also provided iPython notebooks with all of their analysis code.

Meta-analysis All data were indexed against MGI gene symbols (using BioMart mouse version 72), with a total of 31 684 MGI gene symbols having data for at least one platform. The combined effect size was calculated as described for the inverse variance methods described by Choi et al. (33). We calculated effect sizes and significance for the inverse variance meta-analysis based on a random effects model (REM), as this does not assume that there is a single common effect size, but rather a range of true effect sizes with additional sources or variation. Basic calculation and indexing was carried out from the output of AltAnalyze for each study. The analysis was carried out using the PyData stack in the Python programming language (version 2.7.11, as part of the Anaconda python distribution version 4.0.0, from https://www.continuum.io/downloads and is available as a series of interactive notebooks (at https://github.com/LozRiviera/SCN_enrich_Meta, DOI: 10.5281/zenodo.324907). From the meta-analysis, Z-values were used to calculate P-values from two-tailed tests and subsequently to apply a multiple-testing correction in the form of the false discovery rate (FDR) q-value (39). The subsequent FDR-adjusted q-values were calculated from P-values in R (version 3.2.2, (40)), using the ‘qvalue’ package (version 2.0) (41).

Answer 2

[boczniak767]

I'm not an expert, but this two methods differs fundamentally, and uses different measurement scale. I'd rather analyse it separately -possibly using less conservative cut-offs- and look for consistent changes. You should always check the bioconductor packages and publications, maybe I'm wrong.