Skip to main content

Table 3 Correlation coefficients (CCs) between log2IR and six performance metrics plus the average of nine metrics in Table 2 for all four classification algorithms

From: Structure–activity relationship-based chemical classification of highly imbalanced Tox21 datasets

Metrics

Algorithms

RF

RUS

SMO

SMN

F1 score

− 0.7217

− 0.7394

− 0.6941

− 0.9817

MCC

− 0.5778

− 0.6180

− 0.6419

− 0.9761

BA

− 0.6539

− 0.6274

− 0.6227

− 0.9461

AUPRC

− 0.7034

− 0.7148

− 0.8418

− 0.9628

AUROC

− 0.277

− 0.1589

− 0.3713

− 0.7417

SSG

0.7158

0.7072

0.7006

0.9195

Average

− 0.6536

− 0.8421

− 0.7725

− 0.9822

  1. Insignificant CCs are highlighted in bold and are those whose absolute values are smaller than 0.5760, the critical value at α = 0.05 significance level for the degree of freedom df = 10 (i.e., n−2, where n = 12 assays)