Island genome-count sweep
How many genomes does the pangenome island test need? This page subsamples the
293-strain A. fumigatus run (Afumigatus_pangenome / full_v070, 121
representative strains after dereplication) and reruns only the island steps
at each size. It is not a pipeline run: there is no clustering or rescue here.
Result
- Every subset with N <= 20 representative genomes gave 0 islands (0 FDR-significant co-occurring pairs).
- Every subset with N >= 30 gave islands.
- All 7 high-confidence Af293 Starships were hit by islands in every seed at N in {45, 60, 80, 100, 121}.
- The test set
isl45_v1uses the N = 45, seed 3 subset.
| N | seeds | co-occurring pairs | islands | Af293 Starships hit (of 7) | presence matches / scored | island steps (s) |
|---|---|---|---|---|---|---|
| 10 | 3 | 0 | 0 | 0 | 0/1, 0/3, 0/3 | 35 - 66 |
| 15 | 3 | 0 | 0 | 0 | 0/5, 0/4, 0/4 | 91 - 100 |
| 20 | 3 | 0 | 0 | 0 | 0/7, 0/6, 0/4 | 117 - 143 |
| 30 | 3 | 2,273 - 154,865 | 1,546 - 2,306 | 6 - 7 | 4/5, 7/8, 5/7 | 170 - 316 |
| 45 | 3 | 75,194 - 269,098 | 3,152 - 3,570 | 7 | 8/13, 7/11, 10/13 | 324 - 471 |
| 60 | 3 | 47,191 - 530,806 | 3,682 - 4,792 | 7 | 10/12, 12/14, 12/15 | 354 - 616 |
| 80 | 3 | 195,399 - 405,503 | 4,989 - 5,701 | 7 | 14/15, 14/15, 12/13 | 498 - 639 |
| 100 | 3 | 214,515 - 999,062 | 5,957 - 7,045 | 7 | 15/15, 15/16, 15/16 | 591 - 922 |
| 121 | 1 | 1,363,573 | 8,226 | 7 | 16/17 | 1,418 |
Method
- Sampling unit: representative ingroup strains (the co-occurrence test uses only these). Af293 is in every subset; both outgroup references are kept. Seeds 1-3 per N; N = 121 is all.
- Steps (NovInvenio f87fd1e, parameters as in the pipeline): FREQUENCY_BINS, COOCCURRENCE, PAIR_CLASSIFICATION, BUILD_ISLANDS.
- Starship truth: Gluck-Thaler et al. 2025, mBio, doi:10.1128/mbio.01092-25.
- Coordinate check: Table S7, 7 high-confidence Af293 elements. A Starship is hit when an island's Af293 gene lies inside it.
- Presence check: Table S21, 20 Starships. For each Starship, the best Jaccard between an island's strain presence and the Starship's, compared with the 95th percentile of 200 shuffles of the Starship's presence. Starships with fewer than 3 carriers or non-carriers in a subset are not scored.
- Background for the gene fraction: all Af293 genes in non-core families of that subset.
Limits
- Each subset reuses gene families clustered from all 293 genomes. A real N-genome run clusters its own families, so its island counts will differ.
- Runtime here is the island steps only; a full pipeline run also does clustering and rescue.
- N = 121 here gives 8,226 islands;
full_v070(pipeline a394ace) gave 11,775 on the same strains. The cause of the difference was not checked.
Source: studies/fungi/Afumigatus_pangenome/analysis/island_genome_count/ (ISLAND_GENOME_COUNT.md).