How to choose the important clusters for RNAseq data of different groups for downstream analysis/annotation

Question

[alwayshope]

Dear experts,

Could you shed some light on the choice of the clusters of genes in bulk RNAseq analysis?

The mfuzz package can do cluster analysis for groups, while how to choose which cluster is the most important one for the downstream analysis?

Some clusters have a lot of genes that have constant expression but also include a small number of genes that have increased expression. Is there a need to check all the clusters or only focus on the cluster with constantly increased/decreased genes?

Thank you very much!

Answer 1

[rfran010]

I don't really consider myself an expert, but these clusters look strange to me. Did you normalize the data specifically for the clustering Looks like mfuzz may offer this function with

 standardise()

Besides that consideration, you would normally find a cluster that just looksinteresting, and that might include all of them. In a way, as the researcher, you will annotate the clusters. For example, I had 5 clusters and after observing them, I determined there to be 1) early upregulated, 2) late upregulated, 3) transiently upregulted, 4) downregulated, and 5) no change. After that though, what is interesting depends on what your experimental question is.

[alwayshope]

Thanks a lot for your guidance!

I add the standardise() to the eset for the control sample group(average the values of the replicates for each time point). The results are as below. As you can see, some increased genes can be shadowed by the constantly expressed genes, not so obvious as in cluster3, obviously shadowed as in cluster1, I saw the process when generating the cluster plot, the genes increased at 24h actually have very low changes at 5 and 10h, then the constantly expressed lines ploted and covered the lines of the few increased genes.