LncRNA Differential Expression and Analysis of lncRNA and mRNA co-expression.

Question

[conor93]

Hi All,

I am about to undertake RNA-Seq analysis to assess the lncRNA expression induced in viral infected human cells.

Experimental Conditions: I have three conditions that I need to test; wildtype virus infected, mutant virus infected, and uninfected.

I am able to analyse data up to and including differential expression using DESeq2 but i do wish to assess the relationships between differentially expressed human lncRNA and mRNA. The annotation that i use is from GENCODE (Human Primary Assembly) supplemented with lncRNA annotation from LNCipedia.

I have never done co-expression or network analysis before and i am slightly confused as to how to correlate lncRNA expression with that of mRNA expression. I am aware of the R/Bioconductor Package, WGNCA and have attempted to use it with log2 normalized raw counts generated from HTSeq. However, i am not sure as to how to use the result from WGNCA to assess lncRNA and mRNA relationships.

In my last RNA-Seq experiments I used an unstranded NGS library generated from 9 samples (3 replicates per condition as mentioned above) in my analysis, but i am having difficulty in determining some lncRNA gene expression especially those located at intronic and antisense positions. One suggestion was to use a stranded library for future analysis for these lncRNA gene types, though i am unsure.

Questions:

1. What criteria should i select in NGS library preparation to best assess lncRNA differential expression but also to allow for co-expression analysis of lncRNA with mRNA? (strandedness, number of samples, etc)
2. The raw counts generated from HTSeq include both lncRNA genes and protein coding genes using the above annotation. For DESeq2 analysis would it be of any benefit to extract lncRNA genes from the raw-count data and process them separately rather than process everything together through DESeq2 /WGNCA?

Any feedback would be most appreciated. Thanks.

Conor

[Arindam Ghosh]

This seems something similar that I am planning to execute. Could you share some insights from your analysis that I should keep in mind while performing the analysis?

Answer 1

[Kevin Blighe]

Dear Conor, just perform WGCNA with the combined lncRNA and mRNA data, and then look at the derived modules to see which lncRNAs are grouped with which mRNAs. There are major limitations to analyses that are just based on correlation (like WGCNA), though.

You may also consider simply looking at your key lncRNAs and determining their experimentally-confirmed mRNA targets (search Biostars and the Web, generally - there are online databases that have this info), and then see if any of the targets are also key mRNAs in your data. Try to think flexibly in relation to the analysis.

Kevin

[conor93]

Thanks Kevin, will do.

[Kevin Blighe]

Good work, Conor!

[Di Wu]

Dear Prof. Kevin Blighe, I have Agilent microarrays which include mRNA and lncRNA. And use the limma package find many DE mRNA as well as lncRNA. As you said, should I just use limma to find the DE mRNA, and use WCGNA from the expression matrix, them check the DE mRNA from limma and lncRNA in the same module with mRNA? Do you have any pipeline for me to study? Thank you.

[Kevin Blighe]

Dear Di Wu, yes, you can use limma to find the statistically significantly differentially expressed genes (mRNA / lncRNA). For WGCNA, you can use the entire expression matrix, after which you would correlate the modules to your sample metadata / clinical data. There is no need to overlap the WGCNA modules with the statistically significantly differentially expressed genes.

[AS-git]

Hi Kevin Blighe, Thank you for your invaluable responses on Biostar regarding RNAseq analysis. I've gained a lot of knowledge from the insights you've shared. I'm a bit confused about my WGCNA analysis. I've been proceeding with both DE mRNA and DE lncRNA in a single VST normalized matrix for co-expression studies, However, when it comes to module-trait analysis, should I consider only DE mRNA or both?