Skip to main content

Table 2 Performances of DDI-PULearn and the benchmark methods evaluated by 20 runs of 3-fold cross-validation and 5-fold cross-validation

From: DDI-PULearn: a positive-unlabeled learning method for large-scale prediction of drug-drug interactions

EvaluationMethodPrecisionRecallF1-score
3-fold CVVilar’s substructure-based method0.1450.5350.229
 Vilar’s interaction-fingerprint-based method0.3770.5530.447
 Zhang’s weighted average ensemble method0.7820.7030.740
 Zhang’s L1 classifier ensemble method0.7880.7170.751
 Zhang’s L2 classifier ensemble method0.7840.7120.746
 DDI-PULearn0.9020.8220.860
5-fold CVVilar’s substructure-based method0.1450.5350.229
 Vilar’s interaction-fingerprint-based method0.3770.5530.447
 Zhang’s weighted average ensemble method0.7750.6590.712
 Zhang’s L1 classifier ensemble method0.7850.6700.723
 Zhang’s L2 classifier ensemble method0.7830.6650.719
 DDI-PULearn0.9040.8240.862