Spark-GATK

Introduction

Spark-GATK is a genomics analysis framwork based on Apache Spark and ADAM.

In the latest version, WGS pipline is implemented based on this framwork.

Get Started

Build Spark-GATK

Spark-GATK is built using Apache Maven. To build Spark-GATK and its example programs, run:

$ git clone https://github.com/PAA-NCIC/Spark-GATK.git
$ cd Spark-GATK
$ mvn package
$ ...
[INFO] ------------------------------------------------------------------------
[INFO] BUILD SUCCESS
[INFO] ------------------------------------------------------------------------
[INFO] Total time: 12.479 s
[INFO] Finished at: 2016-04-28T09:26:44+08:00
[INFO] Final Memory: 42M/912M
[INFO] ------------------------------------------------------------------------

If you want to run Spark-GATK on a Spark cluster, you should make sure that each machine on your cluster have a copy of this directory.

Installing Spark

A Spark release must be on your system before running Spark-GATK. Our work default to Spark 1.2.0 built for Hadoop 2.4.0, but any more recent Spark distribution should also work. The latest release of Spark can be download from Spark website. More information about installing Spark refers to Spark Installing Document.

Run a example of WGS

A pipline from bam file to vcf file is implemented in Spark-GATK.

Before running a WGS pipline, a human b37 reference and known indel/snp vcf files must be provided in HDFS. The input bam file must be on HDFS too. Then you can write example script like following:

spark_master=spark://master_node_name:7077
gatk_spark_submit=/BASEDIR/NAME/OF/GATK-Spark/bin/gatk-spark-submit

ref_path=/PATH/TO/REFERENCE/human_g1k_v37.fasta                    # must be in HDFS
indel_path=/PATH/TO/KNOWN/_NDEL_FILES/1000G_phase1.indels.b37.vcf  # must be in HDFS
snp_path=/PATH/TO/KNOWN_SNP_FILES/dbsnp_138.b37.vcf                # must be in HDFS

sam_path=/PATH/TO/INPUT_FILE/test.bam                               # input file, must be in HDFS
output=/PATH/TP/OUTPUT_DIR/testResult        # output directory, must be in HDFS

fragment_length=10000
driver_memory=30G
executor_memory=20G
total_executor_cores=20

$gatk_spark_submit haplotype $sam_path $ref_path $output \
   -known_indels $indel_path \
   -known_snps $snp_path \
   -fragment_length ${fragment_length} \
   --master ${spark_master} \
   --driver-memory ${driver_memory} \
   --executor-memory ${executor_memory} \
   --total-executor-cores ${total_executor_cores}

You will see output as following:

Thu Apr 28 12:12:55 CST 2016 Mark duplicates start
Thu Apr 28 12:12:55 CST 2016 Mark duplicates finished
Thu Apr 28 12:12:55 CST 2016 Locally realignment start
Thu Apr 28 12:15:46 CST 2016 Locally realignment finished
Thu Apr 28 12:15:46 CST 2016 Base qualities recalibration start
Thu Apr 28 12:28:31 CST 2016 Base qualities recalibration finished
Thu Apr 28 12:28:31 CST 2016 Haplotype caller start
Thu Apr 28 12:28:57 CST 2016 Haplotype caller Finished, writing result into disk
Thu Apr 28 12:30:57 CST 2016 All tasks finished

Arguments

Spark-GATK support all arguments defined by ADAM, and also defines a series of arguments for WGS pipline.

Arguments for process defination.
 INPUT                                                           : The ADAM, BAM or SAM file to apply the transforms to
 FASTA                                                           : The reference FASTA file to convert
 OUTPUT                                                          : Location to write the transformed data in ADAM/Parquet format
 -mark_duplicate_reads                                           : If remove duplicate reads
 -locallyRealign                                                 : If locally realign indels present in reads.
 -recalibrate_base_qualities                                     : If recalibrate the base quality scores (ILLUMINA only)
 -dump_observations                                              : Local path to dump BQSR observations to. Outputs CSV format.
 -known_snps                                                     : Sites-only VCF giving location of known SNPs
 -knownIndelsFile                                                : VCF file including locations of known INDELs. If not provided, default consensus model will be used
 -max_indel_size                                                 : Maximum length of an INDEL region. (default 500)
 -max_consensus_number                                           : Maximum number of consensus for target region realignment. (default 30)
 -log_odds_threshold                                             : Log-odds threshold for accepting a realignment. (default 5.0)
 -max_target_size                                                : Maximum length of a target region to attempt realigning. (default 3000)
 -fragmentLength                                                 : Maximum fragment length. Default value is 10,000. Values greater than 1e9 should be avoided.
 -partition_size                                                 : Partition size. Values greater than 1e9 should be avoided.(default 1200)

License

Spark-GATK is released under a GNU General Public License.