missing GMM in RNase R Correction

[yiluyucheng]

Hi,

Thanks for developing this amazing tool for circRNA data analysis. According to the 2019 Nature Communications paper, a GMM method was applied to correct RNase R digestion efficiency. However, when reviewing the code, I found GMM was not really involved in the actual correction steps—see lines 68–91 of https://github.com/bioinfo-biols/CIRIquant/blob/master/CIRIquant/coeff.py.

I see that a GMM model is initialized in line 55, but it doesn’t appear to be used afterward.

Could you clarify if I missed something?

Thank you.

[Kevinzjy]

Hi @yiluyucheng,

Thank you for your question! The coeff.py employs a GMM model to fit the distribution of RNase R digestion efficiencies across circRNAs. This distribution is later used as correction factor for further differential expression analysis (https://github.com/bioinfo-biols/CIRIquant/blob/master/CIRIquant/de.py#L156).

The correction of expression levels is performed as illustrated in the Fig.4b of our CIRIquant manuscript. This approach compares the abundance of the same circRNA set in paired total / RNase R samples to get a per-dataset correction constant, which could reduce batch effect and improve accuracy (the lines you mentioned).

[yiluyucheng]

@Kevinzjy
Thanks very much for your prompt response!

If I understand correctly, GMM is only used when calculating DE-score for RNAseq data without replicates, but not involved for correcting reads count.

I saw that coef_mean from coeff.py was mainly used to correct the read counts of BSJ from no RNaseA-treated samples, and obviously coef_mean was not derived from GMM.
And I did not find GMM results were used for DE analysis for samples with replicates.

Please correct me I misunderstood.

Thanks again.

[Kevinzjy]

@yiluyucheng Yes. The differential expression analysis module for replicates does not use RNase R correction, as most cohorts use total RNA-seq (rather RNase R-treated RNA-seq). This aligns with common experimental designs in the field.