Skip to content

v0.42.2 — fast cohort scans: shared construct cache, seeded target-BED depth

Choose a tag to compare

@berntpopp berntpopp released this 19 Oct 20:24
· 10 commits to main since this release
  • perf(cohort): prepare_constructs builds the spliced-alignment cache with the same GTF, padding and shift settings as scan_sample, so scans hit the cache and never load the whole-genome index. The cache refreshes when the GTF changes (#133, #134).
  • perf(depth): the sample median depth for --target-bed uses samtools bedcov -j on a seeded random subsample of target intervals (--target-bed-max-intervals, default 5000, 0 = all; --target-bed-seed, default 42), kept in BED order. This replaces the pure-Python base-level pass over every target.
  • fix: --option=value overrides, an empty GTF, and depth-definition parity (quality_threshold=0).
  • BIH benchmark (real exome BAM, 17 constructs, 8 threads): 4:28 per sample at 173 MB, versus more than 70 min (unfinished) on v0.42.1.