Pileup and coverage

SeQuiLa’s Pileup and coverage benchmarks

Testing environment

Datasets

Type File Size [GB] Download
ES NA12878.proper.wes.md.bam 17 BAM BAI SBI
WGS NA12878.proper.wgs.md.bam 278 BAM BAI SBI
Reference Homo_sapiens_assembly18.fasta 2.9 FASTA FAI

Tools

Software Version Pileup Coverage Multi-threaded Distributed
ADAM 0.36.0 no yes yes yes
GATK 4.2.3.0 yes yes(only intervals) no no
GATK-Spark 4.2.3.0 yes yes(only intervals) yes yes
megadepth 1.1.1 no yes yes(only I/O) no
mosdepth 0.3.2 no yes yes(only I/O) no
sambamba 0.8.1 no yes no no
samtools 1.9 yes yes no no
samtools 1.14 yes yes yes(only coverage, I/O) no
SeQuiLa-cov 0.6.11 no yes yes yes
SeQuiLa 1.0.0 yes yes yes yes

Single node specification

Processor Base freq [GHz] CPUs Total cores Memory [GB] OS Version Disk
Intel(R) Xeon(R) E5-2618L v4 2.20 2 20(40) 256 RHEL 7.8(Maipo) 3TB(RAID1)

Hadoop cluster

Masters Workers Hadoop distribution Total disk HDFS [TB] Total YARN cores Total YARN RAM [TB] Net [Gbits]
6 34 HDP 3.1.4 700 1360(680) 6.8 100

SeQuiLa, ADAM and GATK(Spark) parameters

Parameter Value SeQuiLa only Local-test Cluster-test
spark.biodatageeks.pileup.useVectorizedOrcWriter true yes yes no
spark.sql.orc.compression.codec snappy no yes yes
spark.biodatageeks.readAligment.method hadoopBAM yes yes no
spark.biodatageeks.readAligment.method disq yes no yes
spark.serializer org.apache.spark.serializer.KryoSerializer yes1 yes yes
spark.kryo.registrator org.biodatageeks.sequila.pileup.serializers.CustomKryoRegistrator yes yes yes
spark.kryoserializer.buffer.max 1024m yes yes yes
spark.hadoop.mapred.min.split.size 268435456 yes no yes
spark.hadoop.mapred.min.split.size 134217728 yes yes no
spark.driver.memory 8g no no yes
spark.driver.memory 16g no yes no
spark.executor.memory 4g2 no yes yes
spark.dynamicAllocation.enabled false no yes yes

Other parameters

Software Operation Command line options Multithreading options
ADAM coverage -- coverage -collapse --spark-master local[$n]
GATK pileup PileupSpark --spark-master local[$n]
GATK pileup Pileup –
megadepth coverage --coverage --require-mdz --keep-order --threads $(n-1)
mosdepth coverage -x --threads $(n-1)
sambamba coverage depth base –nthreads=$n3
samtools coverage depth --threads $(n-1)4
samtools pileup mpileup -B -x -A -q 0 -Q 0 –

SeQuiLa

Coverage

import org.apache.spark.sql.{SequilaSession, SparkSession}
val ss = SequilaSession(spark)
ss.sparkContext.setLogLevel("INFO")
val bamPath = "/scratch/wiewiom/WGS/NA12878.proper.wgs.md.bam"
val referencePath = "/home/wiewiom/data/Homo_sapiens_assembly18.fasta"
ss.time{
    ss
    .coverage(bamPath, referencePath)
    .write
    .orc("/tmp/sequila.coverage")
}

Pileup

import org.apache.spark.sql.{SequilaSession, SparkSession}
val ss = SequilaSession(spark)
ss.sparkContext.setLogLevel("INFO")
val bamPath = "/scratch/wiewiom/WGS/NA12878.proper.wgs.md.bam"
val referencePath = "/home/wiewiom/data/Homo_sapiens_assembly18.fasta"
ss.time{
    ss
    .pileup(bamPath, referencePath, true)
    .write
    .orc("/tmp/sequila.pileup")
}

Results single node

Coverage

Pileup

Results Hadoop cluster

Coverage

Pileup

  1. GATK-Spark and ADAM also use fast serialization using Kryo lib and custom registrators. ↩︎

  2. In case of GATK and WGS tests needed to be increased to 8g ↩︎

  3. Does not result in increasing parallelism level ↩︎

  4. Available in version >= 1.13 ↩︎


Last modified February 1, 2023: fix:race condition for local[n] in toCoverage method (#172) (a49c32b)