pone.0244471.g004.tif (126.39 kB)
Download file

AWS WGS benchmarks.

Download (126.39 kB)
figure
posted on 04.02.2021, 18:50 by Charlotte Herzeel, Pascal Costanza, Dries Decap, Jan Fostier, Roel Wuyts, Wilfried Verachtert

The runtime and cost for running the variant calling pipeline on a variety of servers. Similar to the WES experiment, the GATK runs (Java/Intel) show minimal scaling behavior, whereas adding additional CPUs and RAM does improve the runtime when using elPrep. The cost for GATK steeply increases because of its poor scaling whereas the cost for elPrep remains more or less stable across servers. The fastest run with elPrep is 14 times faster than the fastest run with GATK Java mode, and cheaper. The fastest elPrep run is more than 8 times faster than the fastest GATK Intel run, and only 1.4x more expensive. elPrep produces identical outputs to the original haplotype caller algorithm in the GATK Java mode.

History