This repository can be used to recapitulate the development and analysis of a machine learning model approach to somatic variant refinement. DeepSVR contains the raw data (e.g. bam files and manual review labels), code required for data preparation, and validation sets to test the ultimate models. Using the prepared data, we developed three machine learning models (Logistic Regression, Random Forest, and feed-forward Deep Learning). The model that was most consistent with manual revew labels was the Deep Learning model. This model was packaged and is available for use.
A walk-through of the DeepSVR repo can be found on the Wiki page.
Note: Please ensure that you are running these commands using python3 or greater.
1) Clone the DeepSVR GitHub Repo see Repository - Installation
2) Install Anaconda see Downloads - Anaconda
conda config --add channels defaults
conda config --add channels conda-forge
conda config --add channels bioconda
4) Install DeepSVR see BioConda - DeepSVR
conda install deepsvr
deepsvr --help
1) Clone the DeepSVR GitHub Repo see Repository - Installation
docker build -t deepsvr .
docker run deepsvr --help
docker run -v `pwd`:/code deepsvr
Data passed to deepsvr tool needs to be available inside the container. So binding the repo directory path to /code
inside the container allows to access for example training data as /code/wiki_figures/create_classifier/training_data_call.pkl
.