search

SONGDONGYUAN1994 / ClusterDE

A post-clustering differential expression (DE) method robust to false-positive inflation caused by double dipping
https://songdongyuan1994.github.io/ClusterDE/
Other
23 stars 4 forks source link

Using ClusterDE for Cell Type Marker Gene Identification #1

Open altairwei opened 1 year ago

altairwei commented 1 year ago

Hey there,

I've been diving into your ClusterDE tutorial (Perform ClusterDE on a PBMC dataset) and it's pretty cool! I've got a few thoughts and questions I wanted to bounce off you:

  1. Marker Genes: Any tips or tricks on using ClusterDE for spotting cell type marker genes?

  2. 'One-vs-One' vs 'One-vs-Rest': Noticed the tutorial's all about 'one-vs-one' comparisons (like between cluster 2 and 8). But tools like Seurat and Scanpy are all about the 'one-vs-rest' game for markers. How can we play it that way with ClusterDE?

  3. Louvain Algorithm: Trying to get just two clusters with the Louvain algorithm by tweaking the resolution parameter in a 'one-vs-rest' situation seems like a tough gig. Got any hacks or thoughts on that?

Would love to hear your thoughts on these. Thanks for the awesome work and looking forward to your insights!

Cheers, Altair Wei

SONGDONGYUAN1994 commented 1 year ago

Hi Wei, Thank you for your excellent questions! We have considered some of them before, and I will share our thoughts.

  1. spotting cell type marker genes: I am not sure what you mean here. Could you provide a more detailed explanation?
  2. Good question. I think our framework currently cannot be used for 1 vs the others. A simple alternative is to aggregate all 1vs1 markers together in a meaningful way. I will get you back once we have a nice idea.
  3. Yes, I think it is challenging to achieve similar "local clusters" in the "global" setting. Therefore, I think the 1 vs others is not doable with current ClusterDE framework. Please let us know if you have any good thoughts.

Best, Dongyuan

altairwei commented 1 year ago

Hi Dongyuan,

Thanks for your response and for sharing your thoughts!

I apologize for any confusion regarding "Spotting Cell Type Marker Genes". What I meant was, when using ClusterDE, are there specific strategies or methods you'd recommend to effectively identify genes that act as markers for specific cell types? However, I believe your answer to my second point has addressed this.

I'm aware that scran uses a ‘pairwise’ approach for identifying cell type marker genes. This seems like an area where ClusterDE could be particularly useful.

Thanks again for your insights. I'm looking forward to any future updates on ClusterDE!

Best regards, Altair