henhaobio
Sign in

bioRxivBioinformaticsPreprint

The Sentieon Genomics Tools - A fast and accurate solution to variant calling from next-generation sequence data

Donald N. FreediD, Rafael AldanaiD, Jessica A. WeberiD, Jeremy S. EdwardsiD

PDFDOI
In-depth full-text digest available · Jump to digest ↓

Abstract

In the past six years worldwide capacity for human genome sequencing has grown by more than five orders of magnitude, with costs falling by nearly two orders of magnitude over the same period [1], [2]. The rapid expansion in the production of next-generation sequence data and the use of these data in a wide range of new applications has created a need for improved computational tools for data processing. The Sentieon Genomics tools provide an optimized reimplementation of the most accurate pipelines for calling variants from next-generation sequence data, resulting in more than a 10-fold increase in processing speed while providing identical results to best practices pipelines. Here we demonstrate the consistency and improved performance of Sentieons tools relative to BWA, GATK, MuTect, and MuTect2 through analysis of publicly available human exome, low-coverage genome, and high-depth genome sequence data.

In-depth summary

Claude Opus 5.5 · xhigh effort · 17 Oct 2026 · from the full text of version 2 (12 May 2017) · CC BY-ND 4.0

Sentieon's rewrite of GATK and MuTect pipelines runs on average 36x faster than GATK Best Practices on public human data while matching its calls (average F-score 0.9996).

GATK and MuTect are widely used, accurate variant callers, but long runtimes and high memory use for some tasks, plus random downsampling when GATK and MuTect2 run multithreaded, make them costly and not exactly reproducible. Sentieon reimplements the same mathematical models as deterministic, efficiently multithreaded code intended as a drop-in replacement, which matters as sequencing volume and regulatory scrutiny of clinical pipelines grow.

Key results

  1. ShownSentieon DNAseq runs the germline GATK Best Practices workflow about 36 times faster on average, with the largest gains in indel realignment, base-quality score recalibration and HaplotypeCaller.

    Seven whole-exome, 78 low-coverage and two whole-genome samples: average 36x speedup (range 16 to 55x) vs GATK Best Practices; stage speedups 56x (indel realignment), 46x (recalibration) and 30x (HaplotypeCaller). Sentieon's BWA gives identical alignments with an average 1.9x speedup (range 1.0 to 3.9x). Competing tools used built-in multithreading only. · The Sentieon Tools Provide a 10-fold Improvement…

  2. ShownCalls agree with GATK, MuTect and MuTect2 to within the run-to-run variation of those tools, so the Sentieon pipelines can stand in for them.

    RTG Tools vcfeval against the original tools' calls: DNAseq vs GATK average F-score 0.9996 (range 0.9974 to 1.0000); TNseq vs MuTect 0.9994 and vs MuTect2 0.9920; joint genotyping identical to GATK (table not in this text). 'Identical' is defined relative to GATK run-to-run variation reported by Weber et al. · The Sentieon Tools Produce Results Consistent…; Materials and Methods

  3. ShownJoint genotyping with Sentieon GVCFtyper is far faster than GATK GenotypeGVCFs while producing the same genotypes.

    74 gVCFs from the 1000 Genomes Project, chromosome 1 only: 216 s vs 39,555 s, a 183x speedup; GenotypeGVCFs was run with maxAltAlleles 100 for comparability. A whole-genome GATK run was estimated at over two weeks, so only chromosome 1 was tested. · The Sentieon Tools Provide a 10-fold Improvement…; Materials and Methods; The Sentieon Tools Produce Results Consistent…

  4. SuggestedFor tumour-normal somatic calling, the Sentieon TNseq pipelines (TNsnv, TNhaplotyper) are roughly 19 times faster than MuTect2 and 42 times faster than single-threaded MuTect.

    Three tumour-normal pairs: average 19x speedup (range 10x to 24x) vs MuTect2 and 42x (range 38 to 48x) vs MuTect. · The Sentieon Tools Provide a 10-fold Improvement…

  5. ClaimedThe tools are deterministic and never downsample, which the authors say gives perfect run-to-run consistency for medical use; they also claim the implementation can jointly genotype hundreds of thousands of files without intermediate merging.

    Stated as design properties; this text shows no repeated-run comparison for Sentieon itself and no joint-genotyping test beyond 74 gVCFs on chromosome 1. · The Sentieon Genomics Pipeline Tools; Discussion

Methods, data and reproducibility

Runtime and call concordance of the Sentieon DNAseq (germline) and TNseq/TNhaplotyper (tumour-normal) pipelines were compared with BWA-MEM plus GATK Best Practices, MuTect and MuTect2 on public human exome, low-coverage and high-depth genome data, all using built-in multithreading on one dedicated 32-core server; call sets were compared with RTG Tools vcfeval.

  • Public whole-exome samples (germline runtime and concordance benchmark) · human · n = seven
  • Public low-coverage genome samples · human · n = 78
  • Public high-depth whole-genome samples · human · n = two
  • Samples obtained from the precisionFDA platform (their role in each benchmark is not specified in this text) · human · HG001, HG002
  • 1000 Genomes Project gVCF files, chromosome 1, for joint genotyping · human · n = 74
  • Tumour-normal paired samples for somatic calling · human · n = three
Methods & tools
Sentieon DNAseq, Sentieon TNseq (TNsnv, TNhaplotyper), Sentieon DNAseq GVCFtyper, BWA-MEM 0.7.12-r1039, GATK 3.5 (HaplotypeCaller, UnifiedGenotyper, GenotypeGVCFs), MuTect 1.1.5, MuTect2, Picard tools 1.112, SAMtools 1.2, RTG Tools vcfeval 3.5.1
Reagents & environment
Versions: Sentieon 201611, BWA 0.7.12-r1039, Picard 1.112, GATK 3.5, SAMtools 1.2, MuTect 1.1.5, RTG Tools vcfeval 3.5.1; UCSC hg19 reference. Sentieon runs on Linux or OS X without special hardware. Commands are in Supplementary Appendix 1 (not in this text).
Assay conditions & compute
A single dedicated Intel Xeon machine (32 cores at 2.4 GHz, 64 GB RAM, 2TB of striped SSD scratch space) with no other jobs; chromosome-1 joint genotyping took 216 s (Sentieon) vs 39,555 s (GATK).
Reproducibility gaps
The sample list (Supplementary Table 1), per-sample runtime and concordance tables (Supplementary Tables 2-7) and commands are not in this text, and no software download, licence or data links are given.
Limitations (5)
  • Speedups are measured against built-in multithreading only (MuTect is single-threaded and GATK/MuTect2 threading is described as suboptimal); scatter-gather parallelisation used by advanced users was not tested. · stated by the authors
  • Joint genotyping was benchmarked on chromosome 1 only, because a whole-genome GATK run was estimated to take more than two weeks. · stated by the authors
  • Exact agreement is unattainable because multithreaded GATK and MuTect2 downsample randomly and vary between runs, so 'identical' means differences on the scale of that variation. · stated by the authors
  • Accuracy is assessed only as concordance with GATK, MuTect and MuTect2 calls; no comparison with a reference truth set is reported, so the accuracy claim rests on equivalence to those tools. · noted in this summary
  • The benchmark was designed and run by the tool's own developers, and the somatic comparison rests on just three tumour-normal pairs. · noted in this summary

Read this if you run GATK or MuTect germline or tumour-normal pipelines and want to know how much faster a deterministic reimplementation is and how closely its calls agree.

Citations
299

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISKendig '19Supernat '18Zhang '21Zhao '20Wang '24Kumaran '19
78% semantic matchBioinformatics
Katherine I. Kendig, Saurabh Baheti, Matthew A. Bockol+14 more·Frontiers in Genetics·Aug 20, 2019· 270 citations