Moss is a multi-sample somatic single nucleotide variant (SNV) calling tool aiming for discovering variants with a low variant allele frequency that repeatedly appears in several samples. Moss works as an extension to existing single-sample somatic variant callers and improves the recall meanwhile maintains high precision. Moss takes as input the BAM files of multiple samples and corresponding VCF output of the single-sample caller.

1. Installation
   • Using conda (recommended)
   • Build from source (alternative)
     • Dependencies
     • Compilation
   • Docker image
2. Usage instructions
   • Conda/compiled usage
   • Docker usage

1. Create a new conda environment named "moss" and install dependencies:

   conda create -n moss

2. Then activate the created environment: conda activate moss.

3. Install the package into current environment "moss":

   conda install -c bioconda moss

Moss is written in C++11 and thus requires a modern C++ compiler (GCC >= 4.8.1, or Clang). In addition, Moss has the following dependencies.

• HTSlib (>= 1.7)
• CMake (>= 3.9)
• Python (>= 3.6)
• Scikit-allel

mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..

If HTSlib is not in the system path, CMake may not be able to find it. Users then need to manually set the path for htslib using ccmake:

ccmake ..

Then finally make Moss.

make

The compilation results in the executable moss.

The docker image is available on Docker Hub, which can be pulled from Docker Hub by the following command docker pull chuanyiz/moss.

Moss works on top of other somatic variant calling methods, such as Strelka2 and Mutect2. We assume you have already run the base variant caller as their manual suggested and get the VCF files of each sample.

Firstly run the python script scripts/union_candidates.py to generate a VCF file of candidates loci as an input to Moss. For example you can run the toy example in data/:

cd data
python ../scripts/union_candidates.py -f samples.list --normal-name sample0 -t strelka -o candidates.vcf

To run Moss, you need a reference genome FASTA file, BAM files for normal and tumor samples, realigned BAM files (optional but recommended), and a candidate loci VCF. For example, after you've built moss in the build/ directory, you can continue the toy example in data/:

cd build
./moss -r ../data/demo20.fa -b ../data/normal.sort.bam -R ../data/empty.bam -b ../data/sample0.spike.sort.bam -R ../data/empty.bam -b ../data/sample1.spike.sort.bam -R ../data/empty.bam -b ../data/sample2.spike.sort.bam -R ../data/empty.bam -b ../data/sample3.spike.sort.bam -R ../data/empty.bam -l ../data/candidates.vcf -m 4 -t -0.693 --ignore0 --grid-size 200 -o example.vcf

Finally, we filter the result VCF file:

bgzip example.vcf
tabix example.vcf.gz
python ../scripts/post_filter.py --normal-name sample0 -i example.vcf.gz -o example.post_filter.vcf

Organize your data in the same structure as the data/ folder. Modify the config.yaml following the instructions within the file, and it will be used to generate the commands for Moss within the container. When sterting the docker container, you need to bind-mount the data folder by argument -v path:/moss_data:, where path represents the absolute path to your data folder in the host, and /moss_data is the mounted location in the container.

docker run -it -v path:/moss_data: chuanyiz/moss /bin/bash -c "python /moss_scripts/run_moss.py -c /moss_data/config.yaml -o /moss_data/run_moss.sh && bash /moss_data/run_moss.sh"