Fig. 4.

Prediction performance of prediction methods using all six omics combinations in cross-validation. Each classifier is applied to predict the asthma status of children at year 3 using all six omics combinations (in total 63). The missing values of miRNA, mRNA and metabolomics data were imputed using the median values. The heatmap plot shows the average performance of fivefold cross-validations. AdaBoost: A decision-theoretic generalization of online learning and an application to boosting [47]; Bagging: ensemble meta-estimators that aggregate individual predictions to a final prediction [52]; BernoulliNB: Bernoulli Naïve Bayes [55]; GTB: Gradient Tree Boosting [49]; DecisionTree: Decision Trees [53]. Ensemble: aggregate the prediction of all other classifiers together. ERT: An extremely randomized tree classifier [54]. GaussianNB: Gaussian Naïve Bayes. KNN: k-nearest neighbors; LR: Logistic Regression; LRCV: Logistic Regression with build-in validation support to find the optimal parameters; LR-VAE: Logistic Regression with compressed features from VAE (Variational AutoEncoder); LRCV-VAE: Logistic Regression with build-in validation support to find the optimal parameters and compressed feature from VAE (Variational AutoEncoder); MLP: Multi-layer Perceptron; MOGONET: Multi-Omics Graph Convolutional Networks [14]; RF: Random Forest; SVC: Support Vector Classification; Tabnet: Attentive Interpretable Tabular Learning [21]. 1: GWAS; 2: miRNA; 3: mRNA; 4: Microbiome; 5: Metabolomics; 6: DNA methylation