how to get genes and their interactions from RNA-Seq data?

Question

[raza.ul.haq.ruh]

I am working on gene regulatory networks and to construct one I need genes and their interactions with each other. But I can't understand how do I get this information from RNA-Seq data.

[zizigolu]

you need a list of genes in rows and sample names in column as a txt file. the file should include normalized expression values of genes. open Cytoscape, import-network-table then Cyni toolbox, from interface algorithm select one of options for instance ARACNE and run. a network would be created.

[raza.ul.haq.ruh]

and my question is how do I get gene and sample names(their interactions, which I mentioned ) from RNA-Seq data downloaded from SRA NCBI?

[cpad0112]

have a look at string and reactome

Answer 1

[raza.ul.haq.ruh]

Thank you all, I have found a convenient way to resolve this problem by using geneFriends database, you can find find more about it here.

Answer 2

[Kevin Blighe]

Many people do use Cytoscape, as Fereshteh mentions - it is quite good for constructing graphs and networks. I believe that it is now open source (?). Other pathway / network tools, like Ingenuity (commercial), are very comprehensive and professional, and really do a great job.

Another useful implementation, but in R, comes with the igraph package. It is a very broad package and has a lot of functionality. It takes a while to get into it (took me a week initiailly to really grasp it). The first step from an expression dataset to a graph object would be to create a correlation distance matrix and to coerce it to a graph adjacency object (see below).

QUESTION: I don't have a tutorial for this on Biostars but I will put one up if there is demand / interest (?). There are not many tutorials online about it.

graph.adjacency(as.matrix(as.dist(cor(t(MyData), method="pearson"))), mode="undirected", weighted=TRUE, diag=FALSE)

One can also use Euclidean distance

graph.adjacency(as.matrix(dist(MyData, method="Euclidean"))), mode="undirected", weighted=TRUE, diag=FALSE)

Then, by working through numerous (many) functions, you can produce very nice and weird graph objects that you will gaze at for a long time:

pic upload

[Kevin Blighe]

I have created a tutorial for this: Network plot from expression data in R

Answer 3

[svlachavas]

Just to add some complementary comments to the already great answers above. The term "Gene Regulatory Networks", is still a generic term, being part of the general concept of biological networks (which also includes for instance signaling pathways, metabolic networks), which still includes various categories, and numerous approaches for "network-reconstruction". Thus, which is your main goal for inferring gene regulatory networks ? For example:

1) You want to utilize your total expression set of RNA-Seq gene counts, define from this some "co-expressed" modules, and relate them to phenotypic traits or similar downstream analysis, such as functional enrichment ? To see their role in your phenotype pertubation ? Then in R, WGCNA is an excellent choise.

2) Or alternatively, wou would like mostly to infer co-regulatory networks ? That is, infer networks of interacting Transcriptional factors, that regulate a list of DE genes of interest ? and might play a crusial role in your biological system ? Then, CoRegNet R package is a wonderful choise, as it also has an option to use experimentally validated TF-gene interactions, PPIs, etc.

If still you are not interested in R, Cytoscape, Gephi or other standalone tools might be more easy to handle.

Also i would like to suggest a very interesting review about gene regulatory networks, and especially a specific sub-category of these, which is very popular: "gene co-expression networks"

https://academic.oup.com/bib/advance-article/doi/10.1093/bib/bbw139/2888441 (Gene co-expression analysis for functional classification and gene–disease predictions)

Cheers,

Efstathios

[lessismore]

Agree on the amazing review.

Among all the GRN tools, GENIE3 still looks like the most promising one starting from expression data (and i would assume we all want to start from RNA-seq data nowadays). The next challenge in GRNs is to integrate experimentally validated data from multiple experiments (e.g. PPI, CHIP-seq, etc) and eventually combine it which predicted TFBS and interologs in order to have a mixed in silico/experimentally validated most accurate representation of the regulatory context of our process of investigation.

That said, I have a question for @svlachavas and @kevin.
Would you just compile this non-RNA-seq based information (where available for your species of course) or would you still use tools such as the above mentioned CoRegNet, but also cMonkey2 and Merlin+P (which are the only ones able to combine multiple sources of information to assist RNA-seq based GRN inference [even if so far have not been used that much]) ???

Thanks in advance