May I use NB.MClust package for RNASeq genes clustering instead of samples clustering?

Question

[J. Smith]

Hi Friends,

I have find this paper https://bmcbioinformatics.biomedcentral.com/track/pdf/10.1186/s12859-018-2556-9 for RNASeq data clustering (CRAN package: NB.MClust: Negative Binomial Model-Based Clustering). They have used it for samples clustering in RNASeq data. Can it be also possible to clusters genes using this model based clustering?

[ATpoint]

Please explain what kind of data you have. What is genes?

[J. Smith]

I want to use it to cluster RNASeq count data downloaded from TCGA. Genes refers to row names of TCGA count matrix here. I mean, I want to find groups of genes having similar counts across either normal conditions or separately in tumor conditions. I have already separated data for two conditions.

Answer 1

[qian.li10000]

I'm the developer and maintainer of NB.MClust,and just saw your post.

Yes, you can use this package to clustering of genes based on RNA-Seq counts (normalized). However, the assumption in this model-based method requires genes not highly-correlated. I would suggest to filter out genes having strong correlation across all samples (normal+tumor) before using NB.MClust. This method does not require the input of phenotype or condition, so it is completely unsupervised.

I hope this is helpful for your research.

Qian

[J. Smith]

Thanks a lot for your response.

[waiszecandylei]

Hi Qian, what normalization method would you suggest in this case?