Average statistical performance for models with test sets generated using chemical clustering and generated randomly. Clustered statistics are taken from Table 3 and random statistics generated from Table 5. The difference shown is the change in performance when moving from random to clustering^a.

	ACC	MCC	ROC-AUC
Clustered test set	92.2	0.814	0.96
Random test set	92.8	0.832	0.96
Difference	−0.6	−0.018	0

^aACC = accuracy, MCC = Matthews correlation coefficient, ROC-AUC = area under receiver operating characteristic curve.