Explainable deep drug–target representations for binding affinity prediction

BMC Bioinformatics

Table 4 Binding affinity prediction results of the testing set

Method	Protein Rep.	Compound Rep.	\(\downarrow\) MSE	\(\downarrow\) RMSE	\(\uparrow\) CI	\(\uparrow\) \(r^{2}\)	\(\uparrow\) Spearman
Baseline Methods
KronRLS [26]	Smith-Waterman	PubChem-Sim	0.443	0.665	0.847	0.473	0.624
GraphDTA-GCN [31]	1D	Graph	0.315	0.561	0.879	0.625	0.676
GraphDTA-GATNet [31]	1D	Graph	0.307	0.554	0.875	0.634	0.670
SimBoost [27]	Smith-Waterman	PubChem-Sim	0.277	0.526	0.891	0.670	0.694
Sim-CNN-DTA [33]	Smith-Waterman	PubChem-Sim	0.266	0.516	0.884	0.683	0.674
GraphDTA-GIN [31]	1D	Graph	0.255	0.505	0.889	0.696	0.690
GraphDTA-GAT-GCN [31]	1D	Graph	0.254	0.504	0.885	0.697	0.683
DeepDTA [28]	1D	1D	0.222	0.472	0.888	0.735	0.678
DeepCDA [32]	1D	1D	0.202	0.449	0.882	0.760	0.668
Proposed Method
CNN-FCNN	1D	1D	0.177	0.421	0.915	0.789	0.725
Deep Representations Eval.
SVR	CNN Deep Representations		0.203	0.450	0.907	0.759	0.714
GBR	CNN Deep Representations		0.271	0.520	0.894	0.677	0.699
RFR	CNN Deep Representations		0.283	0.532	0.895	0.663	0.703
KRR	CNN Deep Representations		0.453	0.673	0.848	0.461	0.630

Bold indicates that the best performing values associated with each evaluation metric
RFR, random forest regressor; SVR, support vector regressor; GBR, gradient boosting regressor; KRR, kernel ridge regression

ISSN: 1471-2105