Issue
The MERGE_FAMILIES subworkflow passed the seed MSA channel to MERGE_SEEDS as seed_msa.first().
This converts the channel to a value channel holding only the first emission, so every pooled group
of similar families, from every sample, was merged against the first sample’s seed alignments.
For a run with more than one input sample and family merging enabled (the default,
--skip_family_merging false), all merged families of the second and later samples were built from
sequences belonging to the first sample. Downstream GENERATE_FAMILIES outputs derived from those
merged seeds (MSAs, HMMs, recruited sequences and final family assignments) are therefore incorrect
for the affected samples.
Runs with a single input sample, and runs with --skip_family_merging true, are not affected.
Resolution
MERGE_FAMILIES now combines the pooled components with the seed MSA collection by sample id, so each
pooled family is paired with its own sample’s seeds while still allowing multiple groups per sample.
The fix is available in version 2.5.0 onwards (#182).
Users who ran an affected version on a multi-sample samplesheet without --skip_family_merging should
re-run with version 2.5.0 or later.