CardioBoost

This repository includes all the datasets and scripts to reproduce the machine learning training, evaluation, and analysis in the publication:

Zhang X, Walsh R, Whiffin N, et al. Disease-specific variant pathogenicity prediction significantly improves variant interpretation in inherited cardiac conditions. Genet Med. 2021;23(1):69-79. doi:10.1038/s41436-020-00972-3

Dependencies

To install the required R packages: Rscript ./script/install_requiredRpkg.R

Overview of the repository

Variant Classification on Cardiomyopathy and Evaluation

To preprocess the collected features and data splitting (missing values imputation and normalisation): ./script/cardiomyopathy/ml/preprocess_full.Rmd

To train machine learning models using nested cross-validation (CV): optimising hyper-parameters in inner CV loop and evaluating on outer CV loop

Set up how to split dataset in nested CV Rscript ./script/cardiomyopathy/ml/tunecontrol.R

Scripts for training different machine learning models:

• GLMNET: Rscript ./script/cardiomyopathy/ml/cvglmnet.R
• Gradient Boosting: Rscript ./script/cardiomyopathy/ml/gbm.R
• Random Forest: Rscript ./script/cardiomyopathy/ml/rf.R
• AdaBoost: Rscript ./script/cardiomyopathy/ml/ada.R
• K Nearest Neighbours: Rscript ./script/cardiomyopathy/ml/knn.R
• XGBoost (fast implementation of Gradient Boosting): Rscript ./script/cardiomyopathy/ml/xgboost.R
• Bayesian Average Boosting Tree: Rscript ./script/cardiomyopathy/ml/bartMachine.R
• Decision Tree: Rscript ./script/cardiomyopathy/ml/cart.R
• Support Vector Machine with radial basis function kernel: Rscript./script/cardiomyopathy/ml/svm_RBF.R

To evaluate the results of training and select the best trained model: './script/cardiomyopathy/analysis/evaluation_training.Rmd'

To tune and train the selected model: Rscript ./script/cardiomyopathy/ada_tune.R Rscript ./script/cardiomyopathy/train_ada.R

To evaluate the prediction performances on out-of-sample datasets:

• Hold-out test data: './script/cardiomyopathy/analysis/evaluation_holdout_test.Rmd'
• Additional test data: './script/cardiomyopathy/analysis/evaluation_additional_test.Rmd'

To generate predictions for all possible rare missense variants on disease-associated genes: Rscript ./script/cardiomyopathy/ml/cm_prediction_all_possible_variants.R

To evaluate the etiological fraction: './script/cardiomyopathy/analysis/OR_calculation_cardioboost.Rmd' './script/cardiomyopathy/analysis/OR_calculation_mcap.Rmd' './script/cardiomyopathy/analysis/OR_calculation_revel.Rmd'

To evaluate survival outcomes: './script/cardiomyopathy/analysis/survival_analysis.Rmd'

To compare the classification performances on indirectly seen and unseen data: './script/cardiomyopathy/analysis/evaluation_unseen_seen.Rmd'

Variant Classification on Arrhythmia and Evaluation

To preprocess the collected features (missing values imputation and normalisation): ./script/arrhythmia/ml/preprocess_full.Rmd

To train machine learning models using nested cross-validation: optimising hyper-parameters in inner CV loop and evaluating on outer CV loop

Set up how to split dataset in nested CV (i.e., fix the usage of samples in CV loops in the training of different models) Rscript ./script/arrhythmia/ml/tunecontrol.R

Scripts for training different machine learning models:

• GLMNET: Rscript ./script/arrhythmia/ml/cvglmnet.R
• Gradient Boosting: Rscript ./script/arrhythmia/ml/gbm.R
• Random Forest: Rscript ./script/arrhythmia/ml/rf.R
• AdaBoost: Rscript ./script/arrhythmia/ml/ada.R
• K Nearest Neighbours: Rscript ./script/arrhythmia/ml/knn.R
• XGBoost (fast implementation of Gradient Boosting): Rscript ./script/arrhythmia/ml/xgboost.R
• Bayesian Average Boosting Tree: Rscript ./script/arrhythmia/ml/bartMachine.R
• Decision Tree: Rscript ./script/arrhythmia/ml/cart.R
• Support Vector Machine with radial basis function kernel: Rscript ./script/arrhythmia/ml/svm_RBF.R

To evaluate the results of training and select the best trained model: './script/arrhythmia/analysis/evaluation_training.Rmd'

To evaluate the prediction performances on out-of-sample datasets:

• Hold-out test data: './script/arrhythmia/analysis/evaluation_holdout_test.Rmd'
• Additional test data: './script/arrhythmia/analysis/evaluation_additional_test.Rmd'

To generate predictions for all possible rare missense variants on disease-associated genes: Rscript ./script/arrhythmia/ml/arm_prediction_all_possible_variants.R

To compare the classification performances on indirectly seen and unseen data: ' ./script/arrhythmia/analysis/evaluation_unseen_seen.Rmd'