Holmes

Pipeline for structural variation detection from liWGS libraries

Copyright (c) 2017 Ryan L. Collins, Harrison Brand, and the laboratory of Michael E. Talkowski
Massachusetts General Hospital, The Broad Institute, and Harvard Medical School
Contact: Ryan L. Collins [email protected] or Harrison Brand [email protected]

If you use this software, please cite:
Collins RL, Brand H, et al. Defining the spectrum of large inversions, complex structural variation, and chromothripsis in the morbid genome. Genome Biol. (2017)

Code development credits: Ryan L. Collins, Harrison Brand, Matthew R. Stone, Vamsee Pillalamarri, Joseph T. Glessner, Claire Redin, Colby Chiang, Ian Blumenthal, Adrian Heilbut

Table of Contents

1. Overview
2. Modules
   Module 1: Cohort QC
   Module 2: Physical Depth Analysis
   Module 3: Per-Sample Read Pair Clustering
   Module 4: Physical Depth-Based CNV Discovery
   Module 5: Cohort-Wide Joint Read Pair Clustering & Classification
   Module 6: Consensus CNV Categorization
   Module 7: Balanced & Complex SV Resolution
   Module 8: Call Consolidation & Reformatting
   Module 9: SV Annotation
3. Miscellanea & FAQs

Overview

Holmes is a pipeline for SV discovery and annotation in cohorts (n>20-30) of human liWGS libraries. This tool can detect, resolve, and classify all known forms of SV, from canonical copy number variants (CNVs) to balanced rearrangements (e.g. inversions, reciprocal translocations) to more complex chromosomal rearrangements. Currently, this tool only supports human samples and the GRCh37 reference assembly.

The pipeline is run in nine independent modules split into five sequential stages, as shown below:

Note: the codebase for this tool has hard-coded variable paths and other dependencies particular to the Partners Healthcare computing cluster. These dependencies are in the process of being resolved so the tool can be more easily deployed on other clusters. For now, please contact [email protected] with any questions or issues.

Running Holmes

Execution command:

Holmes is invoked by the top-level master script, runHolmes.sh:

runHolmes.sh samples.list parameters_info.sh

Input:

samples.list: three columns, tab delimited
col 1) sample ID
col 2) full path to sample bam
col 3) expected sex (M=XY, F=XX, O=other, U=unknown (will defer to predicted sex by sexcheck))
parameters_info.sh: shell script to export all parameters for pipeline run

Pre-module 1 preparation steps

• Symlinks & indexes all bams
• Creates working and output directory trees
• Loads necessary modules

Module 1: QC

Runs the following:

• Picard EstimateLibraryComplexity
• Picard CollectAlignmentSummaryMetrics
• Picard CollectInsertSizeMetrics
• Picard CollectWgsMetrics
• Samtools flagstat
• Bamtools stats
• Sex Check
• WGS Dosage Bias Check
  Checks for nominal QC values, reports errors to ${OUTDIR}/${COHORT_ID}_WARNINGS.txt
  Writes master QC table to ${OUTDIR}/QC/cohort/${COHORT_ID}.QC.metrics

Module 2: Physical Depth Analysis

• Runs binCov to generate 1kb binned physical depth for each library
• BGZips & tabix indexes each coverage file (for classifier)

Module 3: Per-Sample Read Pair Clustering

• If ${pre_bamstat} isn't set as TRUE, bamstat is run at min cluster size = 3 for each sample
• If ${pre_bamstat}="TRUE", bamstat clusters and stats.file are copied from preexisting paths to ${WRKDIR}
• Removes *pairs.txt and *pairs.sorted.txt to save space

Physical Depth-Based CNV Discovery

• Runs cnMOPS on autosomes on all samples
• Runs cnMOPS on allosomes on samples split by M/F. "Other" sex samples pooled with either M or F depending on value of ${other_assign}
• Merges cnMOPS calls per sample

Module 5: Cohort-Wide Joint Read Pair Clustering & Classification

• Runs classifier
• Patches clusters
• Reclassifies patched clusters
• Applies final classification labels & sets coordinate reporting to be 1st or 3rd quartile of reads, respectively (to avoid overclustering/negative sizes)

Consensus CNV Categorization

• Runs in one of two modes: with or without genotyping information
• Mode chosen by parameter ${min_geno}, set in module6.sh, which corresponds to the minimum number of samples in the cohort to use genotyping

Consensus Groups with Genotyping:

• A [HIGH]: Valid cluster, cnMOPS or genotyping support, <30% blacklist
• B [HIGH]: cnMOPS call, ≥50kb, <30% blacklist, genotyping pass, no clustering overlap
• C [MED]: cnMOPS call, <50kb, genotyping pass, <30% blacklist
• D [MED]: valid cluster, genotyping or cnMOPS support, ≥30% blacklist
• E [MED]: cnMOPS call, ≥50kb, genotyping pass, ≥30% blacklist
• F [LOW]: cnMOPS call, ≥50kb, no clustering support, no genotyping support
• G [LOW]: cnMOPS call, <50kb, genotyping pass, ≥30% blacklist
• H [LOW]: valid cluster, <25kb, no cnMOPS or genotyping support

Consensus Groups without Genotyping:

• A [HIGH]: Valid cluster, cnMOPS support, <30% blacklist
• B [MED]: cnMOPS call, ≥50kb, <30% blacklist, no clustering overlap
• C [MED]: valid cluster, cnMOPS support, ≥30% blacklist
• D [LOW]: cnMOPS call, ≥50kb, ≥30% blacklist
• E [LOW]: valid cluster, <25kb, no cnMOPS support

Returns single merged file each for consensus dels and consensus dups.

Module 7: Balanced & Complex SV Resolution

• Runs inversion classification script
• Runs translocation classification script
• Runs complex linking script
• Runs complex parsing script

Module 8: Call Consolidation & Reformatting

Outputs the following seven variant files:

• Deletion
• Duplication
• Inversion
• Insertion
• Translocation
• Complex
• Unresolved

Module 9: SV Annotation

• Annotates SVs for genic disruption and overlap with ENCODE enhancers, promoters, and TADs