Skip to content

Only one lane for seed - #8

Open
alneberg wants to merge 3 commits into
NationalGenomicsInfrastructure:mainfrom
alneberg:only_one_lane_for_seed
Open

alneberg wants to merge 3 commits into
NationalGenomicsInfrastructure:mainfrom
alneberg:only_one_lane_for_seed

Conversation

@alneberg

Copy link
Copy Markdown
Member

To get demultiplexing to work inside dataflow_testing qwen wanted to strip out all lanes but the first one.

The shipped test flowcell is a 1-lane run (RunInfo.xml LaneCount=1,
SampleSheet.csv declares only lane 1), but the seed document declared
all 8 lanes of the NextSeq 2000. Downstream consumers that plan per
lane (e.g. the demux realm in dataflow_testing) then plan and fail
lane 2-8 branches that have no data on disk.
The seed said Y151;N10;N10;Y151 (4 reads) but the shipped test
flowcell's RunInfo.xml declares two 151-cycle reads (its own
SampleSheet.csv says Y151;Y151), so bcl-convert rejected the rendered
samplesheet with 'The number of reads specified in OverrideCycles
must match the number of reads in the RunInfo.xml'.
@alneberg
alneberg requested a review from fagostini October 20, 2026 12:47
@fagostini

Copy link
Copy Markdown
Contributor

Is it something that makes sense in the mid-/long-term? Meaning, does this branch need merging or is it just for testing until we figure out why it does not work with the additional lanes? We can strip them away, but since the real data on StatusDB look like this, it will then fail in "production".

@alneberg

Copy link
Copy Markdown
Member Author

This is mainly for testing yes. We can leave this as PR and you can have a stab at running the demux with demux realm. I believe bclconvert was complaining for all the other lanes due to the RunInfo.xml file that I had. So its definitely possible to fix in the training data rather than cutting off all other lanes here.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants