Skip to content
Fast fusion detection using kallisto
Branch: master
Clone or download
pmelsted Merge pull request #14 from bgruening/patch-1
Change license and add Bioconda pkg
Latest commit 8617a24 Sep 12, 2017



Fast fusion detection using kallisto


pizzly is a program for detecting gene fusions from RNA-Seq data of cancer samples. It requires running kallisto with the --fusion parameter (available in version 0.43.1 or later).


pizzly uses the SeqAn and requires a recent compiler with c++14 enabled. This means gcc 4.9 or later on linux or clang 3.5 or later. Since these compilers are not always awailable on cluster systems we provide a precompiled binary under releases. Compiling pizzly also requires cmake 3.0.0 or later.

In order to compile just run the following in the source directory

mkdir build
cd build
cmake ..
make install


pizzly can be installed via the Conda package manager from the Bioconda channel.

conda create pizzly --name pizzly -c bioconda


Pizzly requires the reference transcriptome in FASTA format as well as a GTF file describing the transcriptome. We recommend using the Ensembl transcriptomes.

The example below assumes you have your transcriptome in FASTA format as transcripts.fa.gz, the GTF file transcripts.gtf.gz and your paired-end RNA-Seq data sets in r1.fastq.gz and r2.fastq.gz


First we create the kallisto index

kallisto index -i index.idx -k 31 transcripts.fa.gz

Next we quantify using kallisto with fusion detection enabled

kallisto quant -i index.idx --fusion -o output r1.fastq.gz r2.fastq.gz

This creates the file output/fusion.txt which is used by pizzly, finally we run pizzly

pizzly -k 31 --gtf transcripts.gtf --cache index.cache.txt --align-score 2 \
        --insert-size 400 --fasta transcripts.fa.gz --output test output/fusion.txt

The parameters to set are

  • --insert-size, which should be the maximum insert size of the paired-end library (kallisto will estimate this from the reads, default used is 400)
  • --align-score, the number of mismatches allowed when aligning reads to a reference transcript (default used is 2) --ignore-protein, ignore any information about protein coding in the annotation, warning this will probably lead to an increase in the number of false positives reported.
  • --cache, if this file does not exist, pizzly will parse the GTF (which can take up to a minute or two) and store the required data in the cached file. If the cache file exists (if you've run pizzly previously on the same GTF file), pizzly will parse this file instead, which is much faster than parsing the GTF.

A more sophisticated example is in the test directory which contains a snakemake workflow to index, quantify, call fusions and requantify using kallisto and pizzly.


The --output test parameter is used as a prefix and two files are created test.fusions.fasta and test.json, this contains the filtered fusion calls. For unfiltered fusion calls, use the corresponding .unfiltered files.


The scripts subfolder contains useful Python scripts

  • examines an abundance.h5 produced by kallisto and finds the 95th percentile of the fragment length distribution
  • reads the .json output and converts to a simple gene table


pizzly has been tested on Ensembl (versions 75+) and Gencode (version 19+) annotations. We recommend using the latest Ensembl annotations (version 87 GTF, FASTA) for running with pizzly.

Note that for gencode you will need to modify the FASTA file to remove pipe symbols (|) from the target names. The following should work (use gzcat on macosx)

zcat gencode.v26.transcripts.fa.gz  | tr '|' ' ' | gzip -1 >  gencode.v26.transcripts.fixed.fa.gz

The FASTA file used must be the same one that was used to build the kallisto index.



You can’t perform that action at this time.