. 2022 Jun 7;33(5):1741–1753. doi: 10.1007/s11224-022-01975-3

Table 3.

Comparison between classical QSAR models and their corresponding read-across predictions (the best metric values are shown in bold)

Feature combination	Hypothesis	Hyperparameters					$Q_{F 1}^{2}$	$Q_{F 2}^{2}$	MAE	RMSEP
Feature combination	Hypothesis	σ	γ	CTC	Distance threshold	Similarity threshold	$Q_{F 1}^{2}$	$Q_{F 2}^{2}$	MAE	RMSEP
M1	PLSR	-	-	-	-	-	0.831	0.831	0.139	0.179
	RA-ED	1.5	1.5	10	0.5	0.0	0.879	0.878	0.127	0.152
	RA-GK						0.893	0.893	0.121	0.143
	RA-LK						0.909	0.909	0.118	0.132
M2	PLSR	-	-	-	-	-	0.834	0.834	0.157	0.178
	RA-ED	1	1	10	0.6	0.0	0.870	0.870	0.135	0.152
	RA-GK						0.916	0.916	0.121	0.143
	RA-LK						0.911	0.911	0.119	0.132
M3	PLSR	-	-	-	-	-	0.839	0.839	0.154	0.175
	RA-ED	0.75	1.5	10	0.5	0.0	0.862	0.862	0.142	0.162
	RA-GK						0.912	0.912	0.114	0.130
	RA-LK						0.892	0.892	0.132	0.144
M4	PLSR	-	-	-	-	-	0.826	0.826	0.153	0.182
	RA-ED	0.75	1.75	10	0.6	0.0	0.722	0.722	0.163	0.230
	RA-GK						0.931	0.931	0.100	0.115
	RA-LK						0.932	0.932	0.104	0.114