This module implements a method for estimating the continuous fraction of contamination of NGS sequencing reads by parsing relevant information from a VCF file.
pip3 install cython
pip3 install wheel
pip3 install pandas pysam numpy pomegranate scikit-learn
- A VCF file that has been bgzipped with an accompanying tabix index file.
- A float that represents the estimated contamination fraction.
A serialized (pickled) model is used as input into predictSVR.py. This serialized model must have been generated using the exact save version of scikit-learn as that used in predictSVR.py.
The model must first be trained and the model object serialized, after which a VCF can be used as input to generate a contamination prediction.
The training data can be generated as a separate step using a collection of VCFs with known contamination fractions using the following command:
python3 VCFcontam/generate_training_data.py --inputlist <vcflistfile> --loci <bed_file_for_feature_extraction> --feature_out <output_file_name>where the file given to --inputlist is an unheadered two-column file where the
first column is the (full) path to a (bgzipped and tabix-indexed) VCF file with
a known contamination level, and the second column is a float value representing
the known contamination level present in that VCF file. After generating the training
data, the model can be trained using the following command:
python3 VCFcontam/trainSVR.py --training_data <feature_out_from_generate_training_data> --out <serialized_model_output_name>Alternatively, the training data feature extraction and model fitting can all be
performend in a single step by supplying trainSVR.py with a --vcf_metadata flag,
supplying the same two-column tab-delimited file with the paths to the VCFs and
the known contamination level of the VCFs, like so:
python3 VCFcontam/trainSVR.py --vcf_metadata <vcflistfile> --extracted_features <outputfilefortestfeatures> --out <serialized_model_output_name> --loci <bed_file_for_feature_extraction>In the above command, the extracted features are written to an outfile specified by --extracted_features, the serialized model output is written to a file specified by --out.
To estimate contamination from a VCF after training the model, use the following command:
python3 VCFcontam/predictSVR.py --model <trained_model_object_file> --vcf <vcf_file> --index <vcf_file_tbi_index> --loci <bed_file_for_feature_extraction>McCartney-Melstad E, Bi K, Han J, Foo CK. 2021. VCFcontam: A Machine Learning Approach to Estimate Cross-Sample Contamination from Variant Call Data. bioRxiv doi:10.1101/2021.03.12.435007