/** * Biopet is built on top of GATK Queue for building bioinformatic * pipelines. It is mainly intended to support LUMC SHARK cluster which is running * SGE. But other types of HPC that are supported by GATK Queue (such as PBS) * should also be able to execute Biopet tools and pipelines. * * Copyright 2014 Sequencing Analysis Support Core - Leiden University Medical Center * * Contact us at: sasc@lumc.nl * * A dual licensing mode is applied. The source code within this project is freely available for non-commercial use under an AGPL * license; For commercial users or users who do not want to follow the AGPL * license, please contact us to obtain a separate license. */ package nl.lumc.sasc.biopet.pipelines.shiva.variantcallers import nl.lumc.sasc.biopet.core.{ BiopetQScript, Reference } import org.broadinstitute.gatk.queue.QScript /** * Created by pjvan_thof on 11/19/15. */ trait Variantcaller extends QScript with BiopetQScript with Reference { /** Name of mode, this should also be used in the config */ def name: String var namePrefix: String = _ val mergeVcfResults: Boolean = config("merge_vcf_results", default = true) /** * Map of samplename -> (preprocessed) bam file */ var inputBams: Map[String, File] = _ var inputBqsrFiles: Map[String, File] = Map() def init() = {} /** Prio in merging in the final file */ protected def defaultPrio: Int /** Prio from the config */ lazy val prio: Int = config("prio_" + name, default = defaultPrio) /** Final output file of this mode */ def outputFile: File = new File(outputDir, namePrefix + s".$name.vcf.gz") }