Table 3.

Comparison of predictive performance for AUC from three different approaches: RF, FRFL and FRF using CCLE data.

Drug	Correlation					MAE
		KL divergence		Hellinger Distance			KL divergence		Hellinger Distance
	RF	FRFL	FRF	FRFL	FRF	RF	FRFL	FRF	FRFL	FRF
Erlotinib	0.4408	0.4473	0.4620	0.4265	0.4643	0.0546	0.0544	0.0466	0.0552	0.0472
Nilotinib	0.3886	0.4263	0.4601	0.4475	0.5009	0.0465	0.0459	0.0375	0.0457	0.0373
PD-0325901	0.4716	0.5149	0.5775	0.4920	0.5633	0.1353	0.1330	0.1370	0.1352	0.1386
PLX-4720	0.2957	0.3168	0.4308	0.3314	0.4491	0.0494	0.0489	0.0398	0.0492	0.0397
TAE-684	0.2757	0.3245	0.3689	0.2860	0.3337	0.0728	0.0723	0.0688	0.0730	0.0697

For FRFL and FRF, node cost is calculated using f-divergences (KL divergence or Hellinger distance) of the response distributions at 8 different doses. Bold values indicate the best performances.