Table 1.

Results of evaluation tests for different models: mean of MAE and ${\text{Q}}_{\text{F}3}^2$ score obtained from 200 rounds of simulations (the standard deviation is given in parentheses) for the external test sets.

	RM	A	B	C	A*	MF	Ind
MAE	LR	0.47(0.08)	0.51(0.09)	0.60(0.11)	0.43(0.08)	0.51(0.10)	0.39(0.10)
	SVR	0.48(0.09)	0.53(0.10)	0.60(0.11)	0.43(0.08)	0.45(0.11)	0.39(0.11)
	τRAMD	0.76(0.12)	0.39(0.06)	–	0.38(0.08)
	Dummy	0.71(0.11)	0.61(0.11)	0.71(0.11)	0.55(0.14)
${\text{Q}}_{\text{F}3}^2$	LR	0.57(0.21)	0.44(0.30)	0.29(0.30)	0.54(0.23)	0.36(0.52)	0.41(0.52)
	SVR	0.56(0.22)	0.44(0.30)	0.28(0.30)	0.51(0.25)	0.52(0.30)	0.38(0.58)
	τRAMD	−0.41(0.47)	0.69(0.10)	–	0.57(0.23)

Calculations were done for data-sets A, B, and C (see main text) are based on the complete set of 94 compounds. The test sets in these three cases were required to contain some of the outliers found by applying the τRAMD procedure to estimate relative residence times, see Methods for details. A^*–data-set of 80 compounds with outliers discarded. MF—based on molecular property features only. Ind—only IFs of indazole compounds from data-set A are included. For data-set A, the quinazoline compounds (8 compounds) have a mean MAE = 0.60 ± 0.2/0.61 ± 0.2 and ${\text{Q}}_{\text{F}3}^2$ = 0.44 ± 0.4/0.41 ± 0.4 for LR and SVR models, respectively; for the data set MF quinazoline compounds have a mean MAE = 0.59 ± 0.21/0.43 ± 0.25 and ${\text{Q}}_{\text{F}3}^2$ = 0.45 ± 0.39/0.65 ± 0.42 for LR and SVR models, respectively; for the Dummy model ${\text{Q}}_{\text{F}3}^2$ = 0.