Sentieon CLI package version 1.7.1+i, based on upstream Sentieon CLI v1.7.0.
This release accelerates the core dnascope-hybrid path while retaining its public interface and the original final-normalization pipeline:
- direct indexed raw-VCF-to-BED Hybrid selection
- accelerated raw-record Hybrid annotation
- bounded ordered population transfer replacing 493 shard jobs plus concat
- accurate scheduler thread accounting for multithreaded/fused jobs
Matched HG003 direct-CLI result
The release candidate was tested against exact upstream v1.7.0 (1bf377d) using frozen preparations derived from HG003 slim-data 5x Illumina plus 5x ONT FASTQs. Both direct commands used 128 CPUs on i128nvme with Comment=RnD; preparation, queue time, QC, and workflow wrappers were excluded.
| Lane | Slurm job | Direct CLI wall time |
|---|---|---|
| Upstream 1.7.0 baseline | 2467 | 58:05.27 |
| 1.7.1i candidate | 2468 | 41:50.35 |
The candidate was 16:14.92 faster: a 27.97% wall-time reduction and 1.388x throughput. User plus system CPU time fell 6.48%; peak RSS increased 0.67%.
Output validation scanned every decompressed record:
- main hard VCF: 5,661,729 records, exact body equality
- LongReadSV VCF: 15,847 records, exact body equality
- CNV VCF: 2,942 records, exact body equality
- tabix contigs and sampled queries: exact
- header differences: only execution timestamps and job-specific scratch paths in provenance lines
- final normalization: 5,661,729 total and 14,010 realigned records in both lanes
The broader hard-VCF/gVCF multi-sample matrix was not run and is not claimed by this release. Full commands, input hashes, raw GNU-time counters, candidate disposition, and parity criteria are recorded in the 1.7.1i execution ledger.
Validation: 153 tests and 153 doctest-module checks passed, with one expected local-only legacy-bedtools skip that passed in the deployed headnode environment. Black, Flake8, mypy, wheel/sdist build, isolated install, and installed CLI version checks passed.
Fork PR: #1