Skip to main content

Table 3 Performance evaluation of the SVM models trained with various features based on independent testing dataset (3732 ubiquitylation sites and 10,664 non-ubiquitylation sites)

From: UbiSite: incorporating two-layered machine learning method with substrate motifs to predict ubiquitin-conjugation site on lysines

Training features Sensitivity Specificity Accuracy MCC
20D binary coding (20D) 62.59 % 65.85 % 65.00 % 0.253
Amino Acid Composition (AAC) 66.37 % 64.63 % 65.08 % 0.274
Amino Acid Pair Composition (AAPC) 69.05 % 69.05 % 69.05 % 0.340
Position Weight Matrix (PWM) 73.90 % 67.29 % 69.01 % 0.364
Position-Specific Scoring Matrix (PSSM) 73.20 % 68.45 % 69.68 % 0.369
Solvent-Accessible Surface Area (SASA) 63.91 % 61.36 % 62.02 % 0.223
Secondary Structure (SS) 55.60 % 51.34 % 52.45 % 0.061