Doublet Detection in SMART-Seq2 single-cell data
Hi everyone,
I am looking for some feedback on the best practices when it comes to computational identification of doublets in Smart-Seq2 datasets.
I am aware that many tools for doublet detection are available, and when it comes to 10X I generally use both Scrublet (https://github.com/swolock/scrublet) and DoubletFinder (https://github.com/chris-mcginnis-ucsf/DoubletFinder). My understanding, however, is that these tools were developed mainly for droplet-based data, and I am worried that they might perform perfectly in this case...
Probably, all boils down to these questions:
- Given that the expected doublet rate in Smart-Seq2 is generally much lower than for 10X, would it be enough to simply remove cells with a very high number of counts and detected genes (for example, above the median + 3 median absolute deviations) ?
- Should I still run Scrublet and DoubletFinder, adjusting the expected doublet rate to match the expectations for Smart-Seq2?
- Is some manual curation, or a different tool generally recommended?
Thanks a lot in advance!
Luca
• 234 views
•
link
written
by
luca8650 •
0 answers
No answers yet.
Log in to answer this question.
More posts like this
-
Alternatives to Cellbender / barcode selection tools foor poor quality single NUCLEI samples?
written by txema.heredia •
-
After doublet detection in 10x scRNA-Seq data, there is still evidence of doublets
written by yura.grabovska •
-
Tutorial: Single-cell RNA-seq: Annotation: Celltype auto annotation with SCSA
written by Julia Ma •
-
Tutorial: Single-cell RNA-seq: Preprocessing: Data integration and batch correction
written by Julia Ma •
-
Tutorial: Single-cell RNA-seq: Preprocessing: Preprocessing the data of scRNA-seq with omicverse-1
written by Julia Ma •
-
Filter out doublet and Clustering
written by synat.keam •
-
How to choose threshold for scrublet?
written by bioinfo •
-
Tutorial: How to preprocess and visualize beautifully scRNA-seq with omicverse?
written by Starlitnightly •
-
How to adjust for homotypic doublets in the R package DoubletFinder
written by Stevens •
I would just run scDblFinder with the expected doublet rate, and then color the UMAP by the score. Usually, doublets somewhat sit between clusters as scattered outliers. THe score is then usually quite bimodal, so like very few (the true doublets) with scores close to 1, and many with scores like < 0.1, at least in my experience. But generally, as long as you don't have a lot of doublets, their influence on downstream analysis is limited.
Thanks a lot, I am going to give it a try!