Fig. 4.

Performance of various combinations of ML algorithms and sequence-based encoding schemes in predicting human-virus PPIs. Areas under the Precision-Recall curves (AUPRC) show that our pipeline that combined doc2vec embedding and Random Forests (RF) outperforms other combinations, (A) applying 5-fold cross-validation and (B) using an independent test. Considering the computational costs of SVM, note that only half of the whole samples were used to train and assess the SVM classifiers.