GeDi: Generalised suffix array-based Direct SNV caller

Applying suffix arrays to increase the repertoire of detectable SNVs in tumour genomes.

Generalised suffix array-based Direct SNV caller, or GeDi, detects somatic Single Nucleotide Variants (SNVs) within paired tumour-control NGS datasets. GeDi is capable of detecting SNVs reference- and mapping-free. To achieve this, it compares the input sequencing data by construction of multiple suffix arrays from which SNVs can be directly detected; within the arrays, suffixes containing SNVs cluster into intervals enriched with tumour-derived data. GeDi uses mapping only to determine SNV coordinates. To do this, healthy-tissue derived sequences are mapped to the reference genome as a proxy from which SNV coordinates are calculated. Accordingly, not only does GeDi align SNVs reference- and mapping-free, it never aligns tumour data to the reference genome. This enables GeDi to detect SNVs at complex variant loci, such as sites of hypermutation, with high sensitivity.

This work has been published in Coleman, I., Corleone, G., Arram, J. et al. GeDi: applying suffix arrays to increase the repertoire of detectable SNVs in tumour genomes. BMC Bioinformatics 21, 45 (2020). https://doi.org/10.1186/s12859-020-3367-3.

Download

Either clone this repository, or download and unzip the zip file containing GeDi's source code. To clone the GeDi repo, issue the following on the command-line:

git clone https://github.com/izaak-coleman/GeDi

Then move the directory named GeDi (source code directory) to a location you are comfortable with it remaining; once GeDi is compiled, the source code directory cannot be moved without re-compilation (see compilation section below).

Compilation

To reduce the number of dependencies requiring manual installation by the user, we included many of GeDi's dependencies within its source code directory. GeDi will be built directly within its source code directory and as a consequence of the included dependencies, once GeDi is compiled the source code directory cannot be moved without recompilation of GeDi.

Nevertheless, a few dependencies and requirements (listed below) remain to successfully compile GeDi.

• GeDi can be run only on linux machine with x86_64 architecture.
• GNU make must be installed
• cmake must be installed
• g++ version 4.8.1 or greater must be installed
• boost program options library must be installed. On CentOS, install with yum install boost-devel.x86_64, on Ubuntu with sudo apt-get install libboost-all-dev.
• A bowtie2 index of a reference genome must be present. For example, an index of ucsc.hg19.fasta (Human Reference Genome 19) can be built by installing bowtie2 and issuing command bowtie2-build ucsc.hg19.fasta ucsc.hg19.fasta, see bowtie2 reference manual for further details.

Once the dependencies have been installed and GeDi's source code directory is placed in a preferred location, compile GeDi with the following:

cd GeDi
bash install.sh

sudo may be required. Compilation of GeDi will now initiate. Once compilation is complete, test that GeDi compiled successfully by executing

./GeDi --help

which should display GeDi's help message.

Running GeDi on example data

Within GeDi's source code directory is an example_data subdirectory. It will be used in this section to provide an example of how to analyse a paired tumour-control NGS dataset with GeDi in default mode.

Within example_data are the following three files:

• example_dataset.file_list
• example_control_data.fastq.gz
• example_tumour_data.fastq.gz

The tumour and control fastq files form the example paired tumour-control NGS dataset. We assume that whatever fasta preprocessing (e.g quality analysis, de-duplication and trimming) deemed necessary by the user has already been applied. example_dataset.file_list contains the list of fastq file names that form the dataset. example_dataset.file_list will be read by GeDi to load the dataset. Each line of example_dataset.file_list has the structure filename, type, where filename is the absolute path to a fastq file, and type is either C or T, denoting whether the fastq file derives from control or tumour tissue respectively. Hence, example_dataset.file_list contains the following:

./example_data/example_control_data.fastq.gz,C
./example_data/example_tumour_data.fastq.gz,T

Note that relative paths to fastq files are given, but the user should always use absolute paths in their .file_list.

To run GeDi on the example data, issue the following command:

./GeDi -c chr22 -v 30 -t 4 -e 0 -i example_data/example_dataset.file_list -x /path/to/your/bowtie2_index/of/human_ref_genome/ucsc.hg19.fasta -o example_data

Once GeDi has finished executing, it will output three files:

• example_data.SNV_results
• example_data.fastq.gz
• example_data.sam

The .fastq.gz and .sam files contain the healthy-tissue derived sequences mapped to the reference genome as a proxy; they can be deleted. example_data.SNV_results contains the SNV calls made by GeDi for the example dataset. This file has the following content:

Mut_ID	Type	Chr	Pos	Normal_NT	Tumor_NT
0	SNV	chr22	19613299	T	A

When analysing the toy dataset, GeDi detected and called a single SNV on chromosome 22 at position 19613299, with control variant T and tumour variant A. The column headers of an output .SNV_results file describe the following:

• MuT_ID is an arbitrary unique number assigned to each called SNV.
• Type specifies the type of variant GeDi called - current version of GeDi only calls SNVs.
• Chr specifies the chromosome the called SNVs reside.
• Pos specifies the 1-indexed chromosome position of the called SNVs.
• Normal_NT specifies the control variant nucleotide of the called SNVs.
• Tumour_NT specifies the tumour variant nucleotide of the called SNVs.

Running GeDi on user data

By following the above example and making the appropriate substitutions a user will be able to analyse their own data with GeDi.

Accordingly, to run GeDi in default mode, the main substitution is to ensure the .file_list file contains the list of fastq files comprising the users own dataset.

Although issuing ./GeDi --help on the command-line will print information regarding the arguments that must be passed to GeDi's parameters as well as parameter function. Here, we explain what must be passed to the most important of GeDi's parameters (each of which was specified in the above example):

• -i The path and file name of the users .file_list.
• -c If specified, GeDi will filter out all SNV calls that do not reside within the specified chromosome. Note that the passed argument should exactly match the fasta header of the desired chromosome within the reference genome used for alignment. If not specified, GeDi will report all SNVs called regardless of the chromosome they reside in.
• -v Expected or average coverage of the dataset.
• -t Number of threads GeDi will execute with during parallel sections
• -x The location and prefix of the bowtie2 index of Human Reference Genome. Note the argument passed to this parameter should be identical to the argument one would pass to -x if running bowtie2. See bowtie2 reference manual for a further explanation of bowtie2's -x parameter.
• -o Prefix of the output files produced by GeDi. This can include a file path, and GeDi will write files with the prefix at the location specified by the path.
• -e Expected contamination: Proportion of tumour dataset sequencing reads expected to have derived from healthy tissue.

The function of remaining parameters can be found by issuing GeDi --help on the command-line.