Skip to main content

Table 3 Performance of different classification models developed using support vector machine as machine learning technique

From: ProInflam: a webserver for the prediction of proinflammatory antigenicity of peptides and proteins

Feature Thre Sen Spec Acc MCC AUC Parameters
Performance on training data
 AAC 0.6 73.58 70.07 72.92 0.36 0.77 t:2 g:0.005 c:80 j:1
 DPC 0.4 86.11 62.04 81.53 0.45 0.8 t:2 g:0.001 c:10 j:1
 PHY 0.7 91.25 24.82 78.61 0.20 0.57 t:2 g:0.001:c:50:j:4
 DPCHyb_NONE 0.4 87.82 62.04 82.92 0.48 0.84 t:2 g:0.001 c:20 j:1
 DPCHyb_KOOL 0.4 89.54 60.58 84.03 0.49 0.85 t:2 g:0.001 c:4 j:2
 DPCHyb_BETTS 0.3 93.65 62.04 87.64 0.58 0.88 t:2 g:0.001 c:8 j:3
Performance on validation data
 DPCHyb_BETTS 0.3 91.1 50 83.33 0.43 0.71  
  1. The hybrid model prepared using Dipeptide composition based features and MERCI displayed the best performance with an accuracy of 87.6 %. The same model showed an accuracy of 83.3 % on validation dataset