Figure 3.

Performance of NetBCE and otherMLmethods

A. Performances of 84 ML models for the 14 types of features. The AUC values were calculated by five-fold CV. B. ROC curves for NetBCE by different fold CV. C. PR curves for NetBCE by different fold CV. D. Feature representation of the epitopes and non-epitopes using the UMAP method in the input layer of NetBCE. E. Feature representation of the epitopes and non-epitopes in the CNN layer. F. Feature representation of the epitopes and non-epitopes in the BLSTM layer. G. Feature representation of the epitopes and non-epitopes in the attention layer. H. Feature representation of the epitopes and non-epitopes in the fully connected layer. I. Feature representation of the epitopes and non-epitopes in the final classification layer. ML, machine learning; CV, cross-validation; AAC, amino acid composition; CKSAAP, composition of K-spaced amino acid pairs; EAAC, enhanced amino acid composition; ASA, accessible surface area; SS, secondary structure; BTA, backbone torsion angle; AB, AdaBoost; DT, decision trees; KNN, k-nearest neighbors; LR, logistic regression; RF, random forest; SGD, stochastic gradient descent; AUC, area under the receiver operating characteristic curve; ROC, receiver operating characteristic; PR, precision–recall; UMAP, Uniform Manifold Approximation and Projection.