Repository navigation
Performance (#139, #140)
Heavily contaminated samples are much faster:
- Per-base depth: computed with
samtools depthinstead of pysamcount_coverage. On the HPC this took 503 s per call, 3 calls per construct, and now takes 3.3 s. - Junction lookup: binary search over the sorted junction positions in
_scan_plasmid_bam. - Streaming comparison: mate keys and pairing are computed once per read group.
- Local benchmark (1.8 M plasmid reads):
compare_alignments34.2 s → 16.8 s. - Unchanged:
reads_assignment.tsvand the scan counts are byte-identical. - Depth convention: depth now counts
Nbases as covered (samtools convention). On a heavy real sample the relative change is 2.6e-5.
Feature (#138)
--minimap2-threads/ configminimap2_threads: minimap2 threads per alignment.- New default: each of the two concurrent alignments gets
threads // 2. Before, it was 1 thread of 4.