CardioVar: A Machine Learning Framework for Pathogenicity Prediction of Cardiomyopathy Genetic Variants

Figures

• 

  Figure 1. CardioVar development and evolution workflow. Genetic variants were retrieved, annotated, data preparation and feature selection and modeled with several ML algorithms, evaluated using various performance matrices, and then validated by a cardiomyopathy cohort.

• 

  Figure 2. Correlation structure of engineered predictors. Spearman correlations were computed among all numeric features in the ClinVar cardiogene dataset after preprocessing (N = 117 582 variants).

• 

  Figure 3. Holdout performance of five classifiers for variant pathogenicity prediction. (a) Receiver operating characteristic (ROC) curves for gradient boosting (gb), logistic regression (logreg), random forest (rf), RBF-SVM (svm), and XGBoost (xgb) evaluated on the 20% holdout set. (b) Precision–recall (PR) curves on the same holdout set, optimized for class imbalance. (c) Confusion matrices at the default probability threshold of 0.50 for each model.