Single cell cluster evaluation does not output reasonable number of stable clusters
0
0
Entering edit mode
19 months ago
paria ▴ 90

Hello all, I am working on a single nucleus RNA sequencing data of ALS cases and normal controls. I am comparing different regions of brain and spinal cord within ALS patient and between ALS and controls. First, I started to do the integration of the data using SCT+Seurat integration method which result in about 22 clusters. I did evaluate the stability of the clusters using Scclusteval package and hopefully most of the clusters were stable (resolution 0.65, k-param 100, PC 50). But I realized using Seurat integration is overcorrecting my data (I integrated pbmc sample with my dataset and they mixed together after integration). So, I decided to revise integration method and I tried Harmony integration. With harmony when I use scclusteval I'm getting very weird result. I cannot find reasonable number of stable clusters and the number of clusters are significantly lower compared to when I used Seurat integration. For one of my dataset I'm getting 8 clusters with these specification: pc 80, resolution 0.01, param 100 (all 8 are stable) however I expect more sub-clusters and these number could be the major number of clusters. For another dataset I have 5 clusters from which 2 are unstable. I am not sure why I'm getting this different result using these two integration method. I don't know which one I should trust. I really appreciate any comments. Paria

RNA scclusteval single-cell integration harmony sequencing • 1.1k views
ADD COMMENT
1
Entering edit mode

I started to do the integration of the data using SCT+Seurat integration method which result in about 22 clusters

The integration method does not define the number of clusters. If you don't think 22 clusters makes sense, change the clustering resolution (in FindClusters()) to change the number of clusters. For any method, you have to try multiple resolution to fully understand your data.

ADD REPLY
0
Entering edit mode

Thanks and sorry for the late response. I meant that when I use cluster evaluation tools it doesn't show that the clusters are stable. Do you think. I should necessarily used these kinds of tools? and what do you recommend? Thanks, Paria

ADD REPLY
0
Entering edit mode

Stable clusters are not necessarily biologically meaningful and vice versa.

ADD REPLY
0
Entering edit mode

I see. That's a good point. But if I do increase the resolution to get the expected number of clusters how to I make sure that they are stable? do I need to do so? Thanks

ADD REPLY

Login before adding your answer.

Traffic: 1914 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6