PCA to find clusters in gene expression data

Question

[Hashirama]

Dear community,

I finally got some Gene Expression Data on some patients with cancer vs. healthy patients as control. I want to investigate if the expression data form distinct clusters. That's why I performed PCA for dimensionality reduction and plotted it with sns.scatterplot:

You can see that patients with cancers form a cluster (blue) and the healthy patients form another (orange) cluster. But:

1.) Is it valid, to perform PCA to identify cluster or do I need to do other clustering methods, like t-SNE plots or k-means clustering? 2.) How can I show that the clusters are significantly different? Can I calculate p-values? Is it also possible to plot confidence ellipses?

I would be glad for every help!

[4galaxy77]

This is a case where you don’t need a statistical test to find out whether the clusters are distinct.

Answer 1

[ATpoint]

All I see is two distinct populations and one outlier. Remove the outlier sample and compare orange to blue. Does that make sense?

[Hashirama]

Thanks ATpoint for your fast and helpfull answer! But what do you actually mean by "comparing"? What are the next steps?

[ATpoint]

I guess you want sifferential genes? Check tools wuch as limma-voom.