Thanks for your work on DropletUtils — we've relied on emptyDrops for a long time.
We're seeing an issue in a new sciRNA-seq pipeline (switched fromSTAR to STARsolo): for Illumina data, emptyDrops is calling a large proportion of expected cells/barcodes as FALSE. The likely culprit is thealpha parameter — it's much higher and more widely distributed in this case, and manually lowering it restores reasonable results.
Alpha medians across our comparisons:
Illumina, new pipeline (problem case): ~1300, wide spread
Illumina, original pipeline (works fine): ~300, tight
Ultima, new pipeline (works fine): ~150, tight
This rules out both a general "new pipeline" problem and a general "Illumina data" problem — it's specifically Illumina data run through the new STARsolo-based pipeline. The barcode/count matrices are otherwise consistent between pipelines, with only minor UMI count differences (~1–3 UMIs) in the low-UMI range (~100 UMIs), which alone doesn't seem to explain it since Ultima data through the new pipeline is unaffected.
Our question: have you seen emptyDrops behave differently across pipelines when switching STAR to STARsolo? Specifically, could differences in the low-UMI barcode distribution or background profile be inflating the alpha estimate enough to cause this many false calls?
Happy to share example comparisons (original pipeline/Illumina, new pipeline/Illumina, new pipeline/Ultima, and new pipeline/Illumina with manually adjusted alpha) if helpful.
Thank you
Thanks for your work on
DropletUtils— we've relied onemptyDropsfor a long time.We're seeing an issue in a new sciRNA-seq pipeline (switched from
STARtoSTARsolo): for Illumina data,emptyDropsis calling a large proportion of expected cells/barcodes asFALSE. The likely culprit is thealphaparameter — it's much higher and more widely distributed in this case, and manually lowering it restores reasonable results.Alpha medians across our comparisons:
Illumina, new pipeline (problem case): ~1300, wide spread
Illumina, original pipeline (works fine): ~300, tight
Ultima, new pipeline (works fine): ~150, tight
This rules out both a general "new pipeline" problem and a general "Illumina data" problem — it's specifically Illumina data run through the new
STARsolo-based pipeline. The barcode/count matrices are otherwise consistent between pipelines, with only minor UMI count differences (~1–3 UMIs) in the low-UMI range (~100 UMIs), which alone doesn't seem to explain it since Ultima data through the new pipeline is unaffected.Our question: have you seen
emptyDropsbehave differently across pipelines when switchingSTARtoSTARsolo? Specifically, could differences in the low-UMI barcode distribution or background profile be inflating the alpha estimate enough to cause this many false calls?Happy to share example comparisons (original pipeline/Illumina, new pipeline/Illumina, new pipeline/Ultima, and new pipeline/Illumina with manually adjusted alpha) if helpful.
Thank you