Skip to content

Latest commit

 

History

History
134 lines (114 loc) · 7.07 KB

summarize_benchmarking.md

File metadata and controls

134 lines (114 loc) · 7.07 KB
Raw

Benchmarking

Incremental version releases of STITCH have often focused on miscellaneous speed improvements. The purpose of this page is to ensure performance hasn't regressed, as well as to benchmark those speedups across versions and options on a consistent set of data. This is done first using whole chromosome CFW mice data for speed and accuracy and later on a subset of the data to benchmark more options for speed.

The major take aways are

  1. No obvious regression in performance across normal version updates
  2. Consistent gradual speed improvements of approximately 1.5-3X versus earlier releases
  3. Order of magnitude speedups can be obtained for high K using "gridding" approach, with only very slight losses in accuracy

Whole chromosome CFW mice test

This analysis uses 2,073 mice at approximately 0.015X coverage at 152K SNPs using K=4 and 16cores (Intel E5-2690). Performance is assessed using 48 mice on the Illumina MegaMUGA array

Version Options Time (min) Accuracy
1.5.3 output_format='bgen' 29.7 0.973
1.5.3 gridWindowSize=1000 20.8 0.968
1.5.3 gridWindowSize=10000 18.5 0.975
1.5.3 gridWindowSize=100000 17.5 0.96
1.5.3 30.6 0.977
1.5.2 35.1 0.974
1.5.1 34.1 0.977
1.5.0 34.1 0.971
1.4.2 45.3 0.976
1.4.1 45.1 0.975
1.4.0 45.1 0.972
1.3.7 44 0.978
1.3.6 45.5 0.967
1.3.5 45.3 0.968
1.3.4 45.4 0.97
1.3.3 47.6 0.975
1.2.5 61.8 0.971
1.1.1 66.4 0.972

Smaller profiling

These analyses test performance for K=4 or K=20, and for either BAMs or CRAMs, on 1 Mbp (1516 SNPs) of chr19 for 2073 CFW mice at 0.015X. These tests are primarily for speed, given accuracy on the large whole chromosome has been performed above, and showcase speed for a more reasonable choice for outbred samples (K=20), and show that performance on CRAM samples caught up to BAM samples with version >1.3.0.

K=4, BAMs

Version Options Input K Time (min)
1.5.3 output_format='bgen' BAMS 4 2.8
1.5.3 gridWindowSize=1000 BAMS 4 2.1
1.5.3 gridWindowSize=10000 BAMS 4 1.8
1.5.3 gridWindowSize=100000 BAMS 4 1.7
1.5.3 BAMS 4 2.9
1.5.2 BAMS 4 3.5
1.5.1 BAMS 4 3.3
1.5.0 BAMS 4 3.3
1.4.2 BAMS 4 4.1
1.4.1 BAMS 4 4
1.4.0 BAMS 4 3.9
1.3.7 BAMS 4 4.1
1.3.6 BAMS 4 4
1.3.5 BAMS 4 4.1
1.3.4 BAMS 4 4.1
1.3.3 BAMS 4 4.6
1.2.5 BAMS 4 7.1
1.1.1 BAMS 4 6.5

K=4, CRAMs

Version Options Input K Time (min)
1.5.3 output_format='bgen' CRAMS 4 3
1.5.3 gridWindowSize=1000 CRAMS 4 2.3
1.5.3 gridWindowSize=10000 CRAMS 4 2
1.5.3 gridWindowSize=100000 CRAMS 4 1.9
1.5.3 CRAMS 4 3.1
1.5.2 CRAMS 4 3.7
1.5.1 CRAMS 4 3.5
1.5.0 CRAMS 4 3.5
1.4.2 CRAMS 4 4.3
1.4.1 CRAMS 4 4.1
1.4.0 CRAMS 4 4.1
1.3.7 CRAMS 4 4.2
1.3.6 CRAMS 4 4.2
1.3.5 CRAMS 4 4.2
1.3.4 CRAMS 4 4.2
1.3.3 CRAMS 4 4.8
1.2.5 CRAMS 4 13.3
1.1.1 CRAMS 4 NA

K=20, BAMs

Version Options Input K Time (min)
1.5.3 output_format='bgen' BAMS 20 26.2
1.5.3 gridWindowSize=1000 BAMS 20 8.9
1.5.3 gridWindowSize=10000 BAMS 20 4.3
1.5.3 gridWindowSize=100000 BAMS 20 3.2
1.5.3 BAMS 20 26.5
1.5.2 BAMS 20 47.1
1.5.1 BAMS 20 54.5
1.5.0 BAMS 20 54
1.4.2 BAMS 20 60.9
1.4.1 BAMS 20 66.9
1.4.0 BAMS 20 67
1.3.7 BAMS 20 63.1
1.3.6 BAMS 20 59.6
1.3.5 BAMS 20 63.1
1.3.4 BAMS 20 62.9
1.3.3 BAMS 20 57.8
1.2.5 BAMS 20 171.7
1.1.1 BAMS 20 184.2

K=20, CRAMs

Version Options Input K Time (min)
1.5.3 output_format='bgen' CRAMS 20 27
1.5.3 gridWindowSize=1000 CRAMS 20 9
1.5.3 gridWindowSize=10000 CRAMS 20 4.4
1.5.3 gridWindowSize=100000 CRAMS 20 3.3
1.5.3 CRAMS 20 26.8
1.5.2 CRAMS 20 46.3
1.5.1 CRAMS 20 54.5
1.5.0 CRAMS 20 54.7
1.4.2 CRAMS 20 60.7
1.4.1 CRAMS 20 67.3
1.4.0 CRAMS 20 66.9
1.3.7 CRAMS 20 62.8
1.3.6 CRAMS 20 58.5
1.3.5 CRAMS 20 62.9
1.3.4 CRAMS 20 62.8
1.3.3 CRAMS 20 63.5
1.2.5 CRAMS 20 178.4
1.1.1 CRAMS 20 NA