IDEPI performance in predicting phenotypes from genotypes based on training data analyzed previously.

IDEPI metrics were obtained using 5-fold cross-validation. B (balance) is defined as the proportion of "positive" training samples. The number of features (F) was chosen by selecting a value from a pre-defined grid to maximize cross-validation MCC.
1random forests trained on combined sequence and structural features using resistance classifications from the Stanford Drug Resistance Database [51];
2a two-level classifier combining random forest predictions based on an electrostatic hull and hydrophobicity features of the V3 loop (680 features) trained on the same data [27];
3a hierarchical decision tree classifier using composite amino-acid features trained on the same data [35].
4a rule based additive regression model trained to minimize IC50 residuals [45].
5an ensemble classifier using signature rules and logistic regression trained on the same data [44].
IDEPI performance in predicting phenotypes from genotypes based on training data analyzed previously.