Extras¶

Using STAR-fusion to identify endogenous fusions¶

Motivation¶

An important advantage of using RNA-sequencing to identify insertions, rather than targeted DNA-sequencing, is that RNA-seq data provides more information about other events that may be contributing to tumorigenesis. For example, besides the gene-transposon insertions, we may expect somatic gene-fusions (between endogenous genes) to occur in these tumors, which may also contribute to tumorigenesis.

In the IM-Fusion manuscript, we demonstrate this idea for both the SB and the B-ALL datasets by identifying endogenous fusions from STAR RNA-seq alignments using STAR-Fusion. Using this approach, we identified the engineered Etv6-Runx1 fusion in the B-ALL tumors and identified several known oncogenic fusions in the SB tumors.

Here we provide a brief description of how to apply STAR-Fusion to identify endogenous gene fusions. In the first approach, we describe how to run STAR-Fusion as part of IM-Fusion (when STAR is being used to identify insertions). In the second, we describe how to run STAR-Fusion from scratch, which requires extra computation time and disk space but has the advantage of allowing you to use other parameter settings (specifically geared towards identifying endogenous fusions) for STAR during the alignment.

Installing STAR-Fusion¶

STAR-Fusion can be downloaded from the STAR-Fusion GitHub repository. After unpacking, make sure that the STAR-Fusion binary is available in $PATH. Note that STAR-Fusion requires STAR and a few non-standard perl modules to be installed.

Alternatively, STAR-Fusion can also be installed using bioconda:

conda install -c bioconda star-fusion

This should also install any dependencies of STAR-Fusion (such as STAR). Note that the bioconda installation only works on linux at the moment, as bioconda currently does not provide any STAR-Fusion builds for macOS.

Preparing the reference¶

Building a STAR-Fusion reference is quite involved as it requires a number of additional tools and resources to be installed. We therefore recommend using a pre-built reference if possible. Pre-built references are available for human reference genomes and for the mouse mm10 reference genome (using M9/Gencode).

Custom references for other organisms or other reference genome versions can be built using FusionFilter. For more details on how to do so, see the FusionFilter wiki. For convenience, we also provide a Python script that executes the required commands for building a STAR-Fusion reference. Note that this script does expect the required dependencies to be installed (FusionFilter, Repeatmasker and Blast).

Identifying fusions¶

Using IM-Fusion¶

Endogenous gene fusions can be identified using IM-Fusion by supplying a STAR-Fusion reference when identifying insertions using STAR:

imfusion-insertions star
    --fastq sample_s1.R1.fastq.gz \
    --reference references/GRCm38.76.t2onc.star \
    --star_fusion_reference /path/to/star_fusion/reference \
    --output_dir output/sample_s1

This will produce a file called gene_fusions.txt in the output directory, which describes the identified endogenous fusions. See the STAR-Fusion documentation for a detailed description of the structure of this file. Note that the STAR-Fusion reference must use the same reference genome as the IM-Fusion reference (also using the same chromosome naming convention), otherwise no fusions will be found.

From scratch¶

STAR-Fusion can also be run from scratch (starting with fastq files), which has the advantage of performing the STAR alignment with settings specifically tuned for the identification of endogenous gene fusions. Depending on the dataset, this may provide better results than the settings used by IM-Fusion, which are geared towards identifying novel fusion events (with the transposon), rather than focussing on detecting fusions between known reference genes. As a result, analyses based on the IM-Fusion alignments may be more prone to false positives than analyses performed using the STAR settings used by STAR-Fusion.

This type of analysis can be performed using STAR-Fusion with the following command:

STAR-Fusion --genome_lib_dir /path/to/reference \
            --left_fq sample_s1.R1.fastq.gz \
            --right_fq sample_s1.R2.fastq.gz \
            --output_dir ./out/sample_s1

Note that the --right_fq argument is optional and can be omitted for single-end sequencing data.

Alternatively, STAR can first be called separately to perform the alignment, so that you have more control over the supplied options or the location of the generated alignments (which can be useful for further analyses). In this case, the authors of STAR-Fusion recommend using the following parameters for the alignment:

STAR --genomeDir ${star_index_dir} \
     --readFilesIn ${left_fq_filename} ${right_fq_filename} \
     --twopassMode Basic \
     --outReadsUnmapped None \
     --chimSegmentMin 12 \
     --chimJunctionOverhangMin 12 \
     --alignSJDBoverhangMin 10 \
     --alignMatesGapMax 200000 \
     --alignIntronMax 200000 \
     --chimSegmentReadGapMax parameter 3 \
     --alignSJstitchMismatchNmax 5 -1 5 5 \
     --runThreadN ${THREAD_COUNT} \
     --limitBAMsortRAM 31532137230 \
     --outSAMtype BAM SortedByCoordinate

After the alignment, the produced Chimeric.out.junction file(s) can be analyzed to identify fusions with the following command:

STAR-Fusion --genome_lib_dir /path/to/your/CTAT_resource_lib \
            -J Chimeric.out.junction \
            --output_dir star_fusion_outdir