Skip to content
Go to file

NOTE: This is LUMPY 0.2.13 with additional changes to allow lumpyexpress to function when the main file is a CRAM, not a BAM. The splitters and discordants must still be BAM files as LUMPY itself doesn't yet support CRAM as an input. This requires the hexdump command be available.

For questions and discussion about LUMPY please visit the forum at:!forum/lumpy-discuss

Build Status


A probabilistic framework for structural variant discovery.

Ryan M Layer, Colby Chiang, Aaron R Quinlan, and Ira M Hall. 2014. "LUMPY: a Probabilistic Framework for Structural Variant Discovery." Genome Biology 15 (6): R84. doi:10.1186/gb-2014-15-6-r84.

Table of Contents

  1. Quick start
  2. Installation
  3. LUMPY Express usage: Automated breakpoint detection for standard analyses.
  4. LUMPY (traditional) usage: Flexible and customizable breakpoint detection for advanced users.
  5. Example workflows
  6. Test data
  7. Troubleshooting

Quick start

Note that smoove is the recommended way to run lumpy as it collects the best-practices of lumpy and associated tools and will have a shorter run-time and lower false-positive rate than lumpyexpress described below.

Download and install

git clone --recursive
cd lumpy-sv
cp bin/* /usr/local/bin/.

Run LUMPY Express

lumpyexpress \
    -B my.bam \
    -S my.splitters.bam \
    -D my.discordants.bam \
    -o output.vcf



Default method to install:

git clone --recursive
cd lumpy-sv
cp bin/* /usr/local/bin/.

Installing with costom zlib (gzopen64 compile error):

git clone --recursive
cd lumpy-sv
export ZLIB_PATH="/usr/lib/x86_64-linux-gnu/"; #when /usr/lib/x86_64-linux-gnu/
cp bin/* /usr/local/bin/.

LUMPY Express usage

Automated breakpoint detection for standard analyses.

usage:   lumpyexpress [options]

Required arguments

     -B FILE  coordinate-sorted BAM file(s) (comma separated)
     -S FILE  split reads BAM file(s) (comma separated)
     -D FILE  discordant reads BAM files(s) (comma separated)

Optional arguments

-o STR    output [fullBam.bam.vcf]
-x FILE   BED file to exclude
-P        output probability curves for each variant
-m INT    minimum sample weight for a call [4]
-r FLOAT  trim threshold [0]
-T DIR    temp directory [./output_prefix.XXXXXXXXXXXX]
-k        keep temporary files
-K FILE   path to lumpyexpress.config file
            (default: same directory as lumpyexpress)
-v        verbose
-h        show this message


LUMPY Express runs several external program whose paths are specified in scripts/lumpyexpress.config. This config must reside in the same directory as lumpyexpress, or be specified explicitly with the -K flag.

The installation Makefile auto-generates a lumpyexpress.config file and places it in the "bin" directory.


LUMPY Express expects BWA-MEM aligned BAM files as input. It automatically parses sample, library, and read group information using the @RG tags in the BAM header. Each BAM file is expected to contain exactly one sample.

The minimum input is a coordinate-sorted BAM file (-B), from which LUMPY Express extracts splitters and discordants using SAMBLASTER before running LUMPY. Optionally, users may supply coordinate-sorted splitter (-S) and discordant (-D) BAM files which will bypass SAMBLASTER extraction for faster analysis.


LUMPY Express produces a VCF file according to VCF spec 4.2.

LUMPY (traditional) usage

Flexible and customizable breakpoint detection for advanced users.

usage:    lumpy [options]


-g       Genome file (defines chromosome order)
-e       Show evidence for each call
-w       File read windows size (default 1000000)
-mw      minimum weight across all samples for a call
-msw     minimum per-sample weight for a call
-tt      trim threshold
-x       exclude file bed file
-t       temp file prefix, must be to a writeable directory
-P       output probability curve for each variant
-b       output as BEDPE instead of VCF

-sr      bam_file:<file name>,
         id:<sample name>,
         min_mapping_threshold:<mapping quality>,
         weight:<sample weight>,
         min_clip:<minimum clip length>,

-pe      bam_file:<file name>,
         id:<sample name>,
         histo_file:<file name>,
         discordant_z:<z value>,
         min_mapping_threshold:<mapping quality>,
         weight:<sample weight>,

-bedpe   bedpe_file:<bedpe file>,
         id:<sample name>,
         weight:<sample weight>

Example workflows


We recommend aligning data with SpeedSeq, which performs BWA-MEM alignment, marks duplicates and extracts split and discordant read-pairs.

speedseq align -R "@RG\tID:id\tSM:sample\tLB:lib" \
    human_g1k_v37.fasta \
    sample.1.fq \

Otherwise, data may be aligned with BWA-MEM.

# Align the data
bwa mem -R "@RG\tID:id\tSM:sample\tLB:lib" human_g1k_v37.fasta sample.1.fq sample.2.fq \
    | samblaster --excludeDups --addMateTags --maxSplitCount 2 --minNonOverlap 20 \
    | samtools view -S -b - \
    > sample.bam

# Extract the discordant paired-end alignments.
samtools view -b -F 1294 sample.bam > sample.discordants.unsorted.bam

# Extract the split-read alignments
samtools view -h sample.bam \
    | scripts/extractSplitReads_BwaMem -i stdin \
    | samtools view -Sb - \
    > sample.splitters.unsorted.bam

# Sort both alignments
samtools sort sample.discordants.unsorted.bam sample.discordants
samtools sort sample.splitters.unsorted.bam sample.splitters

Running LUMPY

LUMPY has two distinct execution alternatives. LUMPY Express is a simplified wrapper for standard analyses. LUMPY (traditional) is more customizable, for advanced users and specialized experiments.

LUMPY Express
  • Run LUMPY Express on a single sample with pre-extracted splitters and discordants

    lumpyexpress \
        -B sample.bam \
        -S sample.splitters.bam \
        -D sample.discordants.bam \
        -o sample.vcf
  • Run LUMPY Express jointly on multiple samples with pre-extracted splitters and discordants

    lumpyexpress \
        -B sample1.bam,sample2.bam,sample3.bam \
        -S sample1.splitters.bam,sample2.splitters.bam,sample3.splitters.bam \
        -D sample1.discordants.bam,sample2.discordants.bam,sample3.discordants.bam \
        -o multi_sample.vcf
  • Run LUMPY Express on a tumor-normal pair

    lumpyexpress \
        -B tumor.bam,normal.bam \
        -S tumor.splitters.bam,normal.splitters.bam \
        -D tumor.discordants.bam,normal.discordants.bam \
        -o tumor_normal.vcf
LUMPY (traditional)

First, generate empirical insert size statistics on each library in the BAM file

samtools view -r readgroup1 sample.bam \
    | tail -n+100000 \
    | scripts/ \
    -r 101 \
    -X 4 \
    -N 10000 \
    -o sample.lib1.histo

The above script (scripts/ will display mean and stdev to screen. For these examples we will assume the mean is 500 and the stdev is 50.

  • Run LUMPY with paired-end and split-reads.

    lumpy \
        -mw 4 \
        -tt 0 \
        -pe id:sample,bam_file:sample.discordants.bam,histo_file:sample.lib1.histo,mean:500,stdev:50,read_length:101,min_non_overlap:101,discordant_z:5,back_distance:10,weight:1,min_mapping_threshold:20 \
        -sr id:sample,bam_file:sample.splitters.bam,back_distance:10,weight:1,min_mapping_threshold:20 \
        > sample.vcf
  • Run LUMPY on a BAM file with multiple libraries.

    lumpy \
        -mw 4 \
        -tt 0 \
        -pe id:sample,read_group:rg1,bam_file:sample.discordants.bam,histo_file:sample.lib1.histo,mean:500,stdev:50,read_length:101,min_non_overlap:101,discordant_z:5,back_distance:10,weight:1,min_mapping_threshold:20 \
        -pe id:sample,read_group:rg2,bam_file:sample.discordants.bam,histo_file:sample.lib2.histo,mean:500,stdev:50,read_length:101,min_non_overlap:101,discordant_z:5,back_distance:10,weight:1,min_mapping_threshold:20 \
        -sr id:sample,bam_file:sample.splitters.bam,back_distance:10,weight:1,min_mapping_threshold:20 \
        > sample.vcf
  • Run LUMPY on multiple samples with multiple libraries.

    lumpy \
        -mw 4 \
        -tt 0 \
        -pe id:sample1,bam_file:sample1.discordants.bam,read_group:rg1,read_group:rg2,histo_file:sample1.lib1.histo,mean:500,stdev:50,read_length:101,min_non_overlap:101,discordant_z:5,back_distance:10,weight:1,min_mapping_threshold:20 \
        -pe id:sample1,bam_file:sample1.discordants.bam,read_group:rg3,histo_file:sample1.lib2.histo,mean:500,stdev:50,read_length:101,min_non_overlap:101,discordant_z:5,back_distance:10,weight:1,min_mapping_threshold:20 \
        -pe id:sample2,bam_file:sample2.discordants.bam,read_group:rg4,histo_file:sample2.lib1.histo,mean:500,stdev:50,read_length:101,min_non_overlap:101,discordant_z:5,back_distance:10,weight:1,min_mapping_threshold:20 \
        -sr id:sample1,bam_file:sample1.splitters.bam,back_distance:10,weight:1,min_mapping_threshold:20 \
        -sr id:sample2,bam_file:sample2.splitters.bam,back_distance:10,weight:1,min_mapping_threshold:20 \
        > multi_sample.vcf
  • Run LUMPY with low complexity regions excluded.
    Heng Li provides a set of low complexity regions in the supplementary information of his paper, "Toward better understanding of artifacts in variant calling from high-coverage samples" at

    lumpy \
        -mw 4 \
        -tt 0.0 \
        -x btu356_LCR-hs37d5.bed/btu356_LCR-hs37d5.bed \
        -pe bam_file:sample.discordants.bam,,mean:500,stdev:50,read_length:101,min_non_overlap:101,discordant_z:5,back_distance:10,weight:1,id:sample,min_mapping_threshold:1 \
        -sr,back_distance:10,weight:1,id:sample,min_mapping_threshold:1 \
        > sample.exclude.vcf
  • Run LUMPY with regions of very high coverage excluded.
    We can direct lumpy to ignore certain regions by using the exclude region option. In this example we find and then exclude regions that have very high coverage. First we use the script to find the min, max, and mean coverages of the the sr and pe bam files, and to create coverage profiles for both files.

    python ../scripts/ \ \
    #  min:1   max:14  mean(non-zero):2.35557521272
    #  min:1   max:7   mean(non-zero):1.08945936729

    From this output, we will choose to exclude regions that have more than 10x coverage. To create the exclude file we will use the script to create the exclude.bed file

    python ../scripts/ \
        10 \
    exclude.bed \ \

    We now rerun lumpy with the exclude (-x) option

    lumpy \
        -mw 4 \
        -tt 0.0 \
        -x exclude.bed \
        -pe bam_file:sample.discordants.bam,,mean:500,stdev:50,read_length:101,min_non_overlap:101,discordant_z:5,back_distance:10,weight:1,id:sample,min_mapping_threshold:1 \
        -sr,back_distance:10,weight:1,id:sample,min_mapping_threshold:1 \
        > sample.exclude.vcf


SVTyper can call genotypes on LUMPY output VCF files using a Bayesian maximum likelihood algorithm.

svtyper \      
    -B sample.bam \
    -S sample.splitters.bam \
    -i sample.vcf

Test data

The test/ script executes lumpy against several simulated data sets and compares the results to the known correct result. The sample data sets can be found at This tar ball should be extracted into the top-level lumpy directory. The script test/ checks for the the existence of this directory before running LUMPY.


All of the bam files that lumpy processes must be position sorted. To check if your bams are sorted correctly, use the script

python ../scripts/ \
    pe.pos_sorted.bam \
    sr.pos_sorted.bam \
# pe.pos_sorted.bam
# in order
# sr.pos_sorted.bam
# in order
# pe.name_sorted.bam
# out of order:   chr10   102292476   occurred after   chr10   102292893
You can’t perform that action at this time.