Skip to main content
. 2024 May 17;37(6):827–849. doi: 10.1021/acs.chemrestox.3c00352

Table 2. Summary of Performance of State-of-the-Art Models for DTBA Predictiona.

authors name method type data set data set size/interactions target variable R2 rm2 RMSE SCC PCC CI
(116) KronRLS ML Davis 30 056 (68 ligands and 442 targets) Kd 0.580 0.407d 0.573     0.883
0.0482, 4 0.8402,4 0.748e
0.4393,4 0.6603,4 0.861f
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score 0.413 0.3421 0.657     0.7921
0.3272, 4 0.7022,4
0.3633,4 0.6813,4
Metz 35 259 (1.421 compounds and 156 kinases) Ki 0.335   0.781     0.793
0.3282, 4 0.7842,4 0.7362
0.1133,4 0.8993,4 0.6663
(90) SimBoost ML Davis 30 056 (68 ligands and 442 targets) Kd 0.703g 0.6441 0.247     0.884
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score 0.7014 0.6291 0.204     0.847
Metz 35 259 (1421 compounds and 156 kinases) Ki 0.6324   0.116     0.851
(96) PADME Graph-based Davis 30 056 (68 ligands and 442 targets) Kd 0.765   0.429     0.903
0.1442 0.8052 0.7122
0.5913 0.5643 0.8543
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score 0.745   0.433     0.858
0.5092 0.6012 0.7742
0.4713 0.6233 0.7683
Metz 35 259 (1,421 compounds and 156 kinases) Ki 0.665   0.556     0.806
0.4482 0.7122 0.7432
0.3183 0.7903 0.6963
(77) DeepDTA DL Davis 30 056 (68 ligands and 442 targets) Kd   0.630 0.511     0.878
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score   0.673 0.440     0.863
BindingDB 263 534 training samples and 113 142 test samples Ki     0.686h   0.8865  
(117) WideDTA DL Davis 30 056 (68 ligands and 442 targets) Kd     0.512   0.820 0.886
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score     0.423   0.856 0.875
(119) IVPGAN DL Davis 30 056 (68 ligands and 442 targets) Kd 0.945   0.201     0.973
0.8632 0.2892 0.9492
0.9063 0.2203 0.9633
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score 0.766   0.400     0.843
0.6472 0.4702 0.8072
0.7063 0.4493 0.8233
Metz 35 259 (1421 compounds and 156 kinases) Ki 0.628   0.553     0.791
0.6172 0.5482 0.7892
0.5933 0.5743 0.7783
(92) GANsDTA DL Davis 30 056 (68 ligands and 442 targets) Kd   0.653 0.525     0.881
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score   0.675 0.473     0.866
(118) MT-DTI DL Davis 30,056 (68 ligands and 442 targets) Kd   0.665 0.495     0.887
KIBA 118 254 (2,116 drugs and 229 proteins) KIBA score   0.738 0.390     0.882
(93) DGraphDTA Graph-based Davis 30 056 (68 ligands and 442 targets) Kd   0.700 0.450   0.867 0.904
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score   0.786 0.355   0.903 0.904
(120) MONN DL BindingDB 263 534 training samples and 113 142 test samples Ki     0.658   0.895  
(94) GraphDTA Graph-based Davis 30 056 (68 ligands and 442 targets) Kd     0.478     0.893
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score     0.373     0.891
(81) Affinity2Vec Graph-based Davis 30 056 (68 ligands and 442 targets) Kd   0.693 0.490     0.887
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score   0.765 0.352     0.910
(115) DeepMHADTA DL Davis 30 056 (68 ligands and 442 targets) Kd   0.701 0.494     0.895
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score   0.719 0.431     0.876
(121) WGNN-DTA Graph-based Davis 30 056 (68 ligands and 442 targets) Kd     0.456   0.863 0.898
KIBA 118 254 (2116 drugs and 229 proteins) KIBA score     0.360   0.900 0.884
a

Abbreviations: R2: Coefficient of Determination, rm2: Modified r2 ;b RMSE: Root Mean Square Error, SCC: Spearman Correlation Coefficient, PCC: Pearson Correlation Coefficient, CI: Concordance Indexc, ML: Machine Learning, Kd: Dissociation Constant, KIBA: Kinase Inhibitor BioActivity, Ki: Inhibition Constant, DL: Deep Learning.

b

rm2 considers the actual difference between the observed and predicted response data without considering the training set mean.130

c

The CI is a generalization of the AUC and measures the ability of a model to correctly rank the survival times of individuals. It ranges from 0 to 1, where a higher value indicates better predictive performance.131

d

Results retrieved from DeepDTA.77

e

These were obtained under the S2 or cold drug condition; that is, only the drug was not encountered in the training set.

f

These were obtained under the S3 or cold target condition; that is, only the target was not encountered in the training set.

g

Results retrieved from PADME.96

h

Result obtained from Li et al.120