Hi Dr. Ar,

To briefly answer your question, yes you can think of miRNA differential expression analysis as the same as RNA-sequencing, but only as long as you verify a few assumptions as you would be required to do for RNA-seq (e.g. do the data fit a parametric negative binomial distribution?).

I have used DESeq2 in R to analyze miRNA-seq data, to both normalize the miRNA levels within each sample and compute differential miRNA expression. Because of the small sample size within each group, I calculated dispersion estimates via maximum likelihood, assuming that genes of similar average expression strength possess similar dispersion. These gene-wise dispersion estimates were then shrunk according to the empirical Bayes approach, providing normalized count data for miRNA proportional to both the dispersion and sample size. Once the aligned count matrix was normalized, I could then compute differential miRNA expression with these normalized read counts (as described above), the log2(fold-change) according to the Likelihood Ratio Test (LRT), and a Bonferroni-adjusted p value (i.e. q value) were generated for each aligned and annotated gene.

I have attached the R script for you to reference if interested, which outlines part of my miRNA-Seq analysis workflow.

Dear Mark Pepin,

Thank you very much for the explanations. really appreciate it.

Dear Ahmad,

In general, you can use very similar approach as you would use for (long) RNA-Seq. I would just add additional level of filtering to exclude miRNAs with very low raw counts as they are most likely not have any biological relevance a lower the power of your testing.

If you are mostly concerned about the normalization method, you can check following article which tries to to compare some of the most common normalizations (as well as other steps in miRNA-Seq processing) Article Optimization of miRNA-seq data preprocessing

From my experience, TMM (edgeR normalization) and DESeq2 normalization will give you very similar results. Of course, both of them are designed to go well with other steps in their corresponding tools. I would consider edgeR in your analysis as DESeq2 might be too strict when it comes to low-expressed genes and/or if you have small number of replicates. I have seen some analyses where people used Lowess or quantile normalizations for the miRNA expression data.

Hope it helps,

Jan

P.S. There was a great exchange of opinions over the "optimal" miRNA normalization which is quite interesting to read if you are interested. Be aware that the algorithms developed since it was originally published and the results might not be much up-to-date.

Initial publication

Article Evaluation of normalization methods in mammalian microRNA-Seq data

Reaction of edgeR/TMM authors

Article MiRNA-Seq normalization comparisons need improvement

Reaction of authors of the initial publication

Article The poor performance of TMM on microRNA-Seq