Question: number of cell clusters in scRNA-seq
0
4 months ago by
Bogdan580
Palo Alto, CA, USA
Bogdan580 wrote:

Dear all,

considering some of the pipelines that are described in https://bioconductor.org/packages/release/workflows/html/simpleSingleCell.html, would you please let me know, how could I modify the clustering parameters ("the resolution") in order to increase/or to decrease the number of CLUSTERS ?

thanks a lot,

• bogdan
scran scater sc3 • 148 views
modified 4 months ago by Aaron Lun25k • written 4 months ago by Bogdan580
Answer: number of cell clusters in scRNA-seq
1
4 months ago by
Aaron Lun25k
Cambridge, United Kingdom
Aaron Lun25k wrote:

If you're using graph-based clustering methods, increase k to decrease resolution (i.e., bigger and fewer clusters). If you're using hierarchical clustering, you can fiddle with a range of cutreeDynamic parameters, starting with deepSplit. Sometimes dropping minSize is also helpful if you have lots of small clusters that are being forcibly merged to satisfy the minimum size requirement.

Hi Aaron, thanks a lot for the quick reply ! very much appreciate it !

If i may add an additional question, it is related to filtering of cells, and normalization, before performing the clustering :

you'd always recommend data FILTERING (of cells, and genes) before NORMALIZATION (and NOT a NORMALIZATION step before FILTERING), correct ? thanks a lot !

Yes, I would filter out low-quality cells before normalization. Note that computeSumFactors() will automatically filter out genes for you, so that's not necessary; other normalization methods (e.g., library size normalization, spike-in normalization) are robust to low-abundance genes so they don't care.