Differential miRNA Expression using Partek

[leeor]

Hello, I am currently using Partek Genomic Suite Software to analyze smallRNA Seq data for miRNA analysis. I unfortunately do not have biological replicates of my two experimental groups (disease and control) and was wondering what statistical options I have when it comes to detecting differential expression. Partek automatically utilizes an ANOVA, so without any replicates I get a ? for each p-value. I know that going simply on fold-change is not enough so I would like to employ some statistical test to get a better idea of which of my miRNAs are significant.

[aggp11]

Hi,

It'll be very hard to expect any biological significance for any of your detected miRNAs without the presence of biological replicates (if you are planning to publish this data, you'll need biological replicates). Check the guidelines from the ENCODE project regarding the "best practices" for RNA-seq (most ideas should be the same for miRNA-seq) experiments.

http://encodeproject.org/ENCODE/protocols/dataStandards/ENCODE_RNAseq_Standards_V1.0.pdf

As far as getting some p-values from the data you already have (no biological replicates), you could try using DESeq:

http://bioconductor.org/packages/release/bioc/vignettes/DESeq/inst/doc/DESeq.pdf

Just be cautious in making any significant claims about your experiment based on this.

I hope this helps!

[scottbrouilette]

Hi leeor,

In Partek Genomics Suite, during the Quantification step and in the absence of replicates a p-value is calculated for differential expression using a Log Likelihood Ratio test, but this is a very basic statistic as there is no variance component (which comes from having replicates).

Our tool for Differential Expression Analysis uses an ANOVA model which, in the absence of replicates, can calculate ratios and fold-changes, but not p-value as it cannot estimate variance.

If you have no chance to obtain replicates your only option is to perform the Quantification in Genomics Suite, filter the table on p-value (from the Log Likelihood Ratio test) and read count (raw or normalised), and validate expression via qPCR/dPCR or similar.

However at Partek we always recommend the use of true biological replicates as this is the only way to have confidence in your data.

Good luck!

Kind regards,
Scott (European FAS @ Partek)

[leeor]

Thanks for the help! Yes, I did initially plan on having biological replicates but my future steps would involve experimental validation .. so I'll take your suggestions on filtering on the p-value from the Log Likelihood Ratio test.