How does miRdeep2 normalise sequences

Question

[K.patel5]

Hi, I have mirdeep2 output that looks like this.

#miRNA        read_count precursor     total    seq   seq(norm)
mmu-let-7a-5p   43271   mmu-let-7a-1    43271   43271   7658.26
mmu-let-7a-1-3p 784     mmu-let-7a-1    784     784     138.76
mmu-let-7a-5p   43224   mmu-let-7a-2    43224   43224   7649.94

I think using the seq(norm) would be appropriate, but I cannot find how the sequences have been normalised. I think it is something like TPM but am not sure. Does anyone know?

Best, Krutik

Answer 1

[ATpoint]

https://github.com/rajewsky-lab/mirdeep2/blob/master/FAQ.md

How are miRNA expression values reported by the qunatification module normalized?

Expression values are normalized by library size and multiplied by a factor of 1E6 which corresponds technically to counts per million mapped miRNA reads (RPM). The library size is the total number of reads mapping to miRNA precursors.

So a pretty simple and suboptimal normalization. Better use something like edgeR or DESeq2 for a proper normalization, depending on your downstream analysis. What is the analysis goal?

[K.patel5]

I would like to do a standard differential expression analysis.

[ATpoint]

Then I would simple make a count matrix, genes being rows, samples being columns, with the raw counts (read_count) the values, and feed it into e.g. DESeq2. This tool needs raw counts.

[weixuchu63]

Hell ATpoint,

I found your replied is very helpful, thanks. I have another question, what's the criteria to select true-positive miRNAs from all predicted miRNAs? As I understand, it requests significant randfold p-value labeled as yes, high miRDEEP2 score. Do you know a more clear criteria? Thanks. Xu

[ATpoint]

Hi, unfortunately miRNAs are not my expertise, so I cannot comment here. I would suggest you open a new question with all the necessary details, so it gets more attention.

[weixuchu63]

Thanks for your suggestion, just post one.