This repository contains the processing steps required to generate features needed to train new models, to classify a sample as BRCA2d, CDK12d or MMRd and to reproduce the publication. The directory structure includes:

• The data used to train the models and all other supplemental data is located in ./data.
• Figures and associated code from the publicattion are reproduced in ./figures
• Code that was used to find optimal hyperparemeters of each model is located in ./model_training_gridsearchcv
• The data pipeline for predicting a sample can be run through the script ./darcsign_predict.py which calls various functions and files in ./darc_sign_pipeline.

The installation has been tested with python >3.7 and xgboost >1.0 All python dependencies can be installed via conda

#this version is using xgboost=1.5.0 and python=3.9.7
conda create --name darcsign_env --yes numpy scipy pandas matplotlib seaborn scikit-learn pysam xgboost; 

The current version of this pipeline uses the tns and indel features generated by SigProfilerMatrixGenerator which needs to be installed into the created environment:

#activate environment
conda activate darcsign_env;
#Then install SigProfilerMatrixGenerator via pip
pip install SigProfilerMatrixGenerator;
#sigprofiler reference genome also needs to be installed. This tool uses GRCh38
python -c "from SigProfilerMatrixGenerator import install as genInstall; genInstall.install('GRCh38', rsync=False, bash=True)"

Download this repository and the xgboost models on figshare

# clone form get or wget
git clone https://github.com/elieritch/DarcSign.git;
#download and extract model directory from figshare ~65mb
cd DarcSign; 
wget https://figshare.com/ndownloader/files/34540430 -O models.tar.gz; 
tar -xzvf models.tar.gz;
rm models.tar.gz;

In the maf file, the columns needed and the number of the column are:

The names in this table do not matter, the column numbers do. This maf file is converted into a vcf with an awk command called through a subprocess to convert the file into a vcf usable by SigProfilerMatrixGenerator ie..

cmd = f"tail -n +2 {mafpath} | awk \'BEGIN{{FS=OFS=\"\\t\"}} {{print $2,$3,\".\",$5,$6,\".\", \".\", \".\"}}\' >> {vcffile}"
subprocess.call(cmd, shell=True)

One of the products of the sequenza pipeline is a file with the suffix "_segments.txt". This is the other input to the DARC Sign pipeline. Other segmentation data other than Sequenza could also be used but requires the same 4 named columns. The columns that are used from this file are:

The directory also contains a centromere position defining file downloaded from http://hgdownload.cse.ucsc.edu/goldenPath/hg38/database/cytoBand.txt.gz which is used for feature calculation.

seg_file="/path/to/samplename_segments.txt"; #product of sequenza
maf_file="/path/to/samplename.maf"; #product of circuit variant caller or other variant calling tools
output_directory="/path/to/samplename_directory"; #directory will be created if doesnt already exist
sample_name="samplename"; #will be used in figure titles and in the nameing of files 
script="/path/to/darcsign_predict.py"; #from this git repo
#activate appropriate environment
conda activate darcsign_env;
#run the script
python ${script} -m ${maf} -s ${seg} -od ${od} -sn ${sn}

The processed set of features extracted from the input data:

• {samplename}_{kindoffeature}_matrix.tsv which include the raw values of each feature that is used as the input of Darc Sign.
  • ./darc_sign_pipeline/test_data_output/sample_name/sample_name_cnv45feature_matrix.tsv
  • ./darc_sign_pipeline/test_data_output/sample_name/sample_name_snv96feature_matrix.tsv
  • ./darc_sign_pipeline/test_data_output/sample_name/sample_name_ndl83feature_matrix.tsv

Each subdirectory of ./figures contains python scripts and the figures generated by each script. Each figure calls data from the publication which can be found in ./data. The scripts are all run in place using paths relative to their respective hierarchical locations and can be run using the same conda environment as previously installed. eg: the script DarcSign/figures/fig4/predict_and_graph_bc.py produces these graphs and the associated legends.

Figure 5A	Figure 5B