# How should I process the data after removing low-quality cell clusters?

**URL:** <https://discourse.scverse.org/t/how-should-i-process-the-data-after-removing-low-quality-cell-clusters/3855>\
**Category:** scanpy\
**Created:** [November 9, 2025, 7:00pm UTC](https://discourse.scverse.org/t/how-should-i-process-the-data-after-removing-low-quality-cell-clusters/3855 "2025-11-09T19:00:40Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![XJY](https://avatars.discourse-cdn.com/v4/letter/x/a587f6/32.png) [@XJY](https://discourse.scverse.org/u/XJY)\
**Post date:** [November 9, 2025, 7:00pm UTC](https://discourse.scverse.org/t/how-should-i-process-the-data-after-removing-low-quality-cell-clusters/3855/1 "2025-11-09T19:00:40Z")

</div>

* * *

I have already completed the standard single-cell analysis workflow, but I found that one cluster consisted of low-quality cells. I want to remove this cluster, and an AI suggested that I should rerun the preprocessing steps because removing a cluster would affect the identification of highly variable genes and the computation of distances.

> I also want to connect this with subclustering. Based on several tutorials, I plan to do the following:
> 
> ```auto
> raw_adata = adata.raw.to_adata()
> adata_subset = raw_adata[raw_adata.obs['leiden'] != 'CD4 T'].copy()
> 
> sc.pl.highly_variable_genes(adata_subset)
> sc.pp.scale(adata_subset, max_value=10)
> sc.tl.pca(adata_subset)
> sce.pp.harmony_integrate(adata_subset, 'project')
> sc.pp.neighbors(adata_subset, n_neighbors=20, n_pcs=15, use_rep='X_pca_harmony')
> sc.tl.umap(adata_subset)
> sc.tl.leiden(adata_subset, resolution=0.2)
> sc.tl.rank_genes_groups(adata_subset, 
> groupby="leiden_0.2", 
> method="wilcoxon")
> 
> ```
> 
> The AI also mentioned that when running `rank_genes_groups`, I should add the `use_raw` parameter.
> 
> Would this workflow correctly achieve the purpose of reanalyzing the data after removing the low-quality cell cluster?Should I proceed directly from identifying highly variable genes, or should I redo the normalization and log-transformation steps?