Use of Viper with a regulatory network built with my dataset or another one?

Question

[ldetorrente]

Hi,

I am trying to use the viper bioconductor package and having trouble figuring out which regulatory network I should use. Should I build one with aracne-ap directly on my dataset or better to use one from another dataset?

I am working on a breast cancer dataset so my idea was to use viper with the regulatory networks defined in aracne.networks as they were built on TCGA so also cancer related (and even breast cancer related). I have the impression that the example in the vignette on Human B-cell used the same dataset (dset from the bcellViper package) to built the regulatory networks (regulon from the bcellViper package) but I am not 100% sure, does someone know if it's the case?

Also when building regulatory network with Aracne-AP, I have to give it a list of transcription factors. I have seen in the paper “Functional characterization of somatic mutations in cancer using network-based inference of protein activity” that they downloaded 1,813 TF, 969 TCF and 3370 signaling pathway related genes from GO. When you build regulatory networks do you always use all of those? Or do you reduce this list to a smaller one corresponding to TFs you think are important?

Thank you for your help!

Answer 1

[ldetorrente]

The maintainer of the package (Mariano J Alvarez) was actually very fast at answering emails. As maybe other people are wondering I will copy paste his answer here and hope it helps other!

I would definitely use the BRCA network from aracne.networks package (bioconductor). The network should be tissue-matched to the gene expression signatures you want to analyze, but does not need to be generated from the same dataset.

If you plan to run ARACNe, I’d run it using all genes annotated as TFs, neve a subset of some “interesting” TFs. The DPI step in ARACNe requires an unbiased and genome-wide representation of regulators in the network.

[esnolli]

Hi Idetorrente,

I am currently using Viper for trying to predict the TF activity of different datasets with ARACNe generated networks. My ARACNe networks comes from scRNA seq and I want to use bulk RNA seq and microarrays signatures to try to infer TF activity in those samples. However, they differ greatly in terms of gene detection (of course, they were generated with different technologies). What I am doing is to take only common genes found in the network and datasets, and adding zero rows in the datasets for those genes that are uniquely found in the network. I don't know if you had a similar issue and I would like to know how you managed to get around this comparison.

Thank you in advance!

Best,

Esteve