GitHub
09/02/2026, 5:42 AMsourmash_genome_selection.nf.test and species_preference.nf.test were extremely slow because 7-8 selected genomes per test each needed a full real Prokka annotation (no pre-computed GFF is possible for genomes fetched live from NCBI at test time).
The 3 species kept (Metallosphaera javensis, Aeropyrum pernix, Ignisphaera cupida) were chosen specifically to preserve every --species_preference (local/completeness/gtdb) and --genomeset_mode (joint/sample) behaviour the original 7-species set exercised -- see the new README section for the full reasoning and the actual CheckM/GTDB completeness numbers behind each choice. All 3 reuse genomes, GFFs, and GTDB metadata already in this dataset; nothing new was downloaded from NCBI.
Also fixes a design smell in the original dataset: its three read samples were built from entirely disjoint species, so --genomeset_mode joint vs sample was never actually exercised against a genome appearing in more than one sample. The three new samples deliberately share species pairwise instead.
testdata/archaeal_duplicates.gtdbtk.tsv (all 7 species) is left untouched, since nf-core/magmap's test_bakta profile still uses it via bakta_test.genomes.csv, which references all 6 local_* genomes directly. Everything else from the original 7-species set is removed as now-orphaned.
All matches were verified with real `sourmash sketch`/`prefetch` runs before committing (not just assumed): every sample's simulated reads cleanly match both their public-representative and local-genome signatures, well above the pipeline's Sourmash Gather threshold.
Companion PR on nf-core/magmap switching `test_sourmash_genome_selection`/`test_species_preference` to this new dataset will follow once this merges.
Generated with some help from Claude Code.
nf-core/test-datasetsGitHub
09/02/2026, 6:57 AMGitHub
09/02/2026, 7:06 AM