Simulated microarray expression data

Question

[Jan-Niklas]

Dear all,

I want to compare the performance of some gene set analysis methods and therefore want to simulate my own expression data to overcome the lack of a gold standard. The simulated data should be a good approximation of real biological data with it's complex characteristics and distributions. Genes should be modeled as known correlated blocks, which than can be identified by gene set analysis methods and detection rates can be estimated.

I found the Umpire R package Link, which looks promising, but an annotation of which gene sets are up and down regulated seems to be missing. Does anybody have experience working with Umpire or know a different tool for this purpose or a paper which describes the workflow to simulate expression data?

With best regards,

Jan-Niklas

[lchaba]

Hello Jan,

Did you manage to figure this out? I have just installed Umpire package but I'm not sure how to go about it. Kindly help. I'll appreciate.

Answer 1

[Benn]

"Ain't nothing but the real thing..." - Marvin Gaye

There is a huge database called GEO full with the real stuff, why not try it with real data.

[Jan-Niklas]

Heard about that. :) But for all these datasets the ground truth is unknown. Sure, you can pick datasets which study a specific phenotype and you could assume that pathways associated with this phenotype show a significant correlation. But after all I would like to simulate data with known ground truth.

Answer 2

[cpad0112]

For future reference, try :

1. Ruvcorr package and corresponding simulation page: https://rdrr.io/bioc/RUVcorr/man/simulateGEdata.html
2. sgnesR package: download from GitHub (not in CRAN and bioconducor repositories). No installation instructions on GitHub (as of 15th June, 2018)- https://github.com/shaileshtripathi/sgnesR