Workflows for data pre-processing and initial calling of somatic SNP, Indel, and copy number variants optimized by Intel for on-premises infrastructure

This repository contains the modified WDLs for Somatic workflow optimized to run on BIGstack.

a. Set the correct path in SomaticPairedSingleSampleWf.wdl for where your datasets reside. b. SomaticPairedSingleSampleWf.wdl uses Samtools Sort instead of Picard Sort. c. If BWA threads for SamToFastqAndBwaMemAndMba stage has to be changed, make the change in two places: SomaticPairedSingleSampleWf.wdl CommonTasks.wdl d. mutect2.wdl - Enabled Mutect2 with pairHMM and smith-waterman AVX implementations.

Assuming the environemnt has been setup to offload the pairhmm kernel of Mutect2 to FPGA - the below changes must be enabled in the mutect2.wdl file (based on the comments) to make use of the FPGA.

a. In the WDL file, for task M2 runtime section, uncomment the line: require_fpga: "yes". Also modify the -pairHMM flag to use EXPERIMENTAL_FPGA_LOGLESS_CACHING.

Use the Download_Somatic_Data.sh script to download all the datasets needed for this workflow.

For on-prem, this workflow uses dockerized tools. However, if mutect2 uses FPGA, we recommend modifying the task M2 to use tools from the filesystem to work with the FPGA. There's a bug running FPGA with docker.

This workflow works with Cromwell v30.2 and HTCondor backend. Going <30.2 or >30.2 has caused issues.