Differential expression analysis (DEA) of lncRNA

Question

[bart]

Hi all,

I'm relatively new to RNA-seq analysis and want to perform differential expression analysis (DEA) of lncRNAs. I have read multiple different protocols and I think the following method would be valid and I hoped to receive feedback on it:

• calculate RPKM values from the raw counts from htseq-count
• remove genes with 50th-percentile RPKM =0, and then only keep the genes with 90th-percentile RPKM < 0.1 as suggested here (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4777353/)
• perform normalization using EdgeR's TMM method
• use lncDIFF (https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-019-5926-4) package to select normalized genes with log 2 fold change >1 and BH FDR <0.05. This package should have better performance compared to other differential expression tools such as DESeq2, edgeR, etc.

Would this be a good way to perform DEA of lncRNAs?

Thanks!

Answer 1

[Gordon Smyth]

Statistical analysis of RNA-seq for lncRNAs is done the same as for any other RNA-seq analysis and there isn't any particular need for new pipelines. I strongly recommend using a regular RNA-seq pipeline with read counts instead of RKPMs. You can follow the regular edgeR documentation. edgeR is quite capable of using fractional counts, if that is what you have. I don't personally agree that lncDIFF has better performance than edgeR. I would tend to analyse the lncRNAs together with protein-coding genes in order to make global normalization methods like TMM more reliable.