Method#
Why the numbers are what they are. Every page here exists because a first-pass answer was wrong and the correction needed writing down – these are not derivations for their own sake, they are the measurements the defaults are set from.
page |
the question it answers |
|---|---|
how long is my barcode really, and how much of it is the same base in every molecule |
|
how often is a barcode base called wrong, and can I believe the answer |
|
is the barcode long enough for this library, and how many molecules collided |
|
do the reads assigned to one molecule belong to one molecule |
|
the reads of one 10x molecule are not co-terminal, so what is a consensus of them |
|
the highest quality a consensus is allowed to claim, and why it is not Q60 |
|
what “significant” means for a split, when both margins have to be preserved |
|
the spike-in metric, and what it says the pipeline gets right |
- UMI statistics
- Barcode error against depth
- Barcode space, collisions and the error budget
- Grouping accuracy: Calib, UMI-tools, fgbio
- Fragmented libraries — why 10x needs a different consensus
- The RT/PCR floor — the quality no consensus can beat
- Three permutation nulls
- Validation — the spike-in metric