BenchmarkDR is a modular end-to-end pipeline to train and evaluate machine learning prediction models on genomic short-read sequencing data. The predicted phenotype can be either a binary classification (e.g. susceptible to a drug/ resistant) or a regression (e.g. minimum inhibitory concentration (MIC)).
Clone this repo and set up an environment containing Snakemake
git clone https://github.com/WGS-TB/BenchmarkDR.git
conda activate base
mamba create -c conda-forge -c bioconda -n snakemake snakemakeTo run the pipeline on your own data, you have to structure your input data (bacterial read and labels) in following way:
your_bacterium_1 | |---- data (place paired-end short reads here) | |---- reference | | | | | reference_for_your_bacterium_1.fasta | |---- AllLabels.csv To adapt configuration of the pipeline, there are different config files. It is necessary to adapt the config in the main folder: |-- workflow | |--rules | | |-- prediction | | |-- models_config | | |-- ... | |-- ... | |-- config |-- Snakefile
PATH_DATA: "path to folder containing your bacterial folders"
BACTERIA: ["your_bacterium_1"]
OUTPUT_DIR: "path to folder where you would like to save all files created by the pipeline"
REPRESENTATION: ["gene_presence_absence", "snps"] # "kmer"
MODE: "MIC" # "Classification"
DRUGS: ["drug_1", ...] Column names of in your AllLabels.csv file
METHODS: ["sklearn_LinR", "sklearn_LinR_l1", "sklearn_LinR_l2", "sklearn_LinR_elasticnet", "sklearn_SVMR", "sklearn_GBTR", "sklearn_RFR"] #Classification ["sklearn_LR_l1", "sklearn_LR_l2", "sklearn_LR_elasticnet", "sklearn_SGD_l1", "sklearn_SGD_l2", "sklearn_SGD_elasticnet", "sklearn_DT", "sklearn_RFC", "sklearn_ET", "sklearn_ADB", "sklearn_GBTC", "sklearn_GNB", "sklearn_CNB", "sklearn_SVM_l1", "sklearn_SVM_l2", "sklearn_KNN", "INGOT"]
OPTIMIZATION: "None" # 'GridSearchCV', 'RandomizedSearchCV' and "None"snakemake -s Snakefile --use-conda