Skip to main content

Table 4 Performances of selectivity modeling using the two-step A1AR-A2AAR difference model or the selectivity-window model

From: Quantitative prediction of selectivity between the A1 and A2A adenosine receptors

Model ClassMCCSensitivitySpecificityPPVNPVROC
Trained on all double points, tested on all double points (non-binders and semi-selective compounds always true/false negative)
 A1AR-A2AAR differenceA1AR0.15 ± 0.130.17 ± 0.140.99 ± 0.000.18 ± 0.120.98 ± 0.010.76 ± 0.09
A2AAR0.11 ± 0.070.19 ± 0.090.94 ± 0.020.15 ± 0.070.94 ± 0.010.67 ± 0.14
Dual0.26 ± 0.050.76 ± 0.020.50 ± 0.070.68 ± 0.020.59 ± 0.030.66 ± 0.02
 Selectivity-windowA1AR0.07** ± 0.090.030.030.99 ± 0.000.10** ± 0.100.97 ± 0.000.87 ± 0.03
A2AAR0.22 ± 0.120.15 ± 0.090.99 ± 0.000.42 ± 0.180.94 ± 0.010.74 ± 0.07
Dual0.36 ± 0.060.85 ± 0.030.48 ± 0.030.69 ± 0.010.70 ± 0.060.70 ± 0.02
Trained on all double points, tested on only A1AR, A2AAR, and dual class
 A1AR-A2AAR differenceA1AR0.19 ± 0.160.17 ± 0.140.99 ± 0.000.28 ± 0.180.96 ± 0.010.75 ± 0.09
A2AAR0.28* ± 0.120.19 ± 0.090.98 ± 0.010.48* ± 0.150.91 ± 0.010.72 ± 0.15
Dual0.25 ± 0.070.76 ± 0.020.57 ± 0.110.91 ± 0.020.28 ± 0.040.70 ± 0.05
 Selectivity-windowA1AR0.17** ± 0.230.03 ± 0.030.99 ± 0.010.50** ± 0.500.92 ± 0.010.88 ± 0.04
A2AAR0.32* ± 0.150.15 ± 0.091.00 ± 0.000.75* ± 0.250.82 ± 0.030.80 ± 0.07
Dual0.33 ± 0.110.84 ± 0.040.46 ± 0.060.80 ± 0.020.55 ± 0.100.66 ± 0.04
Trained on only A1AR, A2AAR, and dual class, tested on only A1AR, A2AAR, and dual class
 Selectivity-windowA1AR− 0.05*** ± 0.000.00 ± 0.000.99 ± 0.010.00*** ± 0.000.92 ± 0.010.81 ± 0.05
A2AAR0.23 ± 0.180.25 ± 0.120.94 ± 0.030.43 ± 0.220.83 ± 0.030.66 ± 0.11
Dual0.04 ± 0.120.69 ± 0.050.35 ± 0.090.73 ± 0.030.32 ± 0.090.53 ± 0.08
  1. The class indicates the performance for that particular selectivity class: A1AR-selective, A2AAR-selective, and dual (non-selective). The query compounds were categorized based on post-classification of the selectivity predictions: A1AR-selective when pActivity difference ≥ 2, A2AAR-selective when pActivity difference ≤ − 2, and dual binder when pActivity difference ≥ − 1 and ≤ 1
  2. MCC Matthews Correlation Coefficient, PPV positive predictive value, NPV negative predictive value, ROC receiver operating characteristic
  3. *1/5 folds failed, **3/5 folds failed, ***4/5 folds failed