Summary of results for various DNN architectures for several targets in initial investigations. Best performing networks on the test data are highlighted in red. Full results can be found in the ESI (Tables S5–S9). The first column represents the NN architecture, showing the number of neurons in each hidden layer^a.

	Training					Validation					Test
	SE	SP	ACC	MCC	ROC-AUC	SE	SP	ACC	MCC	ROC-AUC	SE	SP	ACC	MCC	ROC-AUC
AChE
[10]	88.7	83.9	86.6	0.726	0.93	84.9	80.7	83.1	0.655	0.90	84.2	78.9	81.9	0.631	0.89
[100]	90.7	88.4	89.7	0.791	0.96	87.4	83.2	85.6	0.706	0.92	86.2	80.7	83.8	0.670	0.90
[1000]	88.0	83.7	86.2	0.718	0.93	85.5	78.0	82.3	0.637	0.89	84.4	78.8	82.0	0.632	0.88
[10,10]	90.7	89.7	90.3	0.802	0.96	86.1	82.9	84.7	0.688	0.92	84.3	82.4	83.5	0.664	0.90
[100,100]	91.5	91.3	91.4	0.826	0.97	87.1	85.2	86.3	0.721	0.92	85.0	84.2	84.7	0.689	0.91
[1000,1000]	95.2	96.6	95.8	0.915	0.99	88.0	86.7	87.4	0.744	0.93	84.7	84.0	84.4	0.684	0.92
ADORA2A
[10]	97.6	89.9	95.0	0.888	0.98	97.2	90.2	94.7	0.884	0.98	97.2	88.5	94.2	0.871	0.97
[100]	97.8	92.9	96.1	0.913	0.99	96.9	90.9	94.8	0.886	0.98	97.2	90.2	94.8	0.884	0.98
[1000]	97.5	90.7	95.2	0.893	0.98	97.2	89.5	94.6	0.879	0.98	97.0	89.1	94.3	0.872	0.97
[10,10]	97.8	92.7	96.0	0.911	0.99	97.6	90.6	95.3	0.893	0.98	97.0	90.0	94.6	0.880	0.98
[100,100]	98.1	93.7	96.6	0.924	0.99	96.8	90.8	94.8	0.883	0.98	96.9	90.5	94.7	0.881	0.98
[1000,1000]	99.0	77.8	91.7	0.817	1.00	97.3	92.4	95.6	0.903	0.98	96.7	91.2	94.8	0.884	0.98
AR
[10]	58.0	99.3	88.3	0.691	0.88	59.1	98.9	88.3	0.691	0.87	55.8	99.0	87.5	0.667	0.86
[100]	69.1	98.7	90.9	0.759	0.91	64.4	98.1	89.1	0.711	0.87	64.5	98.3	89.3	0.715	0.86
[1000]	65.0	98.6	89.7	0.727	0.89	61.6	98.2	88.5	0.693	0.86	61.5	98.3	88.6	0.695	0.86
[10,10]	67.1	99.0	90.5	0.750	0.90	62.7	98.5	89.0	0.708	0.86	61.6	98.6	88.8	0.701	0.87
[100,100]	76.1	99.4	93.2	0.823	0.95	69.2	97.8	90.2	0.740	0.87	68.0	98.1	90.1	0.737	0.87
[1000,1000]	73.3	99.4	92.5	0.804	0.94	65.8	97.9	89.3	0.717	0.87	64.4	98.2	89.2	0.713	0.87
hERG
[10]	93.5	53.5	77.5	0.529	0.87	91.6	48.2	74.3	0.454	0.82	92.0	46.1	73.7	0.441	0.81
[100]	94.1	49.9	76.4	0.508	0.86	92.2	45.8	74.2	0.443	0.81	92.9	44.1	73.4	0.438	0.80
[1000]	89.7	64.3	79.7	0.568	0.87	84.6	59.5	72.7	0.458	0.82	87.0	55.0	74.2	0.450	0.81
[10,10]	94.1	85.0	90.5	0.800	0.97	86.1	67.0	78.4	0.545	0.86	86.3	63.8	77.3	0.519	0.85
[100,100]	96.2	90.5	93.9	0.873	0.98	84.9	69.8	78.8	0.555	0.86	85.1	65.5	77.3	0.519	0.84
[1000,1000]	95.0	87.5	92.0	0.833	0.98	84.2	66.8	77.2	0.520	0.86	83.4	65.5	76.2	0.498	0.84
SERT
[10]	99.2	72.4	93.4	0.799	0.98	99.1	66.1	91.7	0.752	0.97	99.0	67.6	92.1	0.760	0.97
[100]	99.0	89.2	96.9	0.906	0.99	98.4	83.8	95.1	0.856	0.98	98.6	83.1	95.2	0.857	0.98
[1000]	99.2	77.2	94.4	0.831	0.98	98.8	73.8	93.3	0.797	0.97	99.1	73.9	93.5	0.805	0.97
[10,10]	99.0	89.7	97.0	0.909	0.99	98.9	82.1	95.1	0.857	0.98	98.7	83.1	95.3	0.858	0.98
[100,100]	99.4	95.8	98.6	0.959	1.00	98.2	86.1	95.6	0.867	0.98	98.6	86.8	96.0	0.882	0.99
[1000,1000]	99.4	98.2	99.1	0.975	1.00	98.1	91.1	96.5	0.897	0.99	98.4	90.5	96.6	0.901	0.99

^aSE = sensitivity, SP = specificity, ACC = accuracy, MCC = Matthews correlation coefficient, ROC-AUC = area under receiver operating characteristic curve.