Machine-Learning Prediction of Oral Drug-Induced Liver Injury (DILI) via Multiple Features and Endpoints

Table 3

Paired -test results of AUC values during 10-fold cross-validations between NSap1 and NSap2 as negative examples.

		Logistic regression		Random forest
Database	Features
DrugDex	Fatal hADRs	-3.80	7.69-04	-2.83	7.53-03
	Liver failure	-3.33	2.46-03	-1.51	1.40-01
	Liver transplantation	-2.33	2.63-02	-2.50	1.69-02
	Jaundice	-3.10	4.04-03	-3.69	1.01-03
	Biomarker increase	-2.76	9.05-03	-0.59	5.60-01
	Hepatomegaly	-0.35	7.28-01	-0.72	4.77-01
	Hepatitis	-3.15	3.52-03	-3.00	4.70-03
	All hADRs	-0.12	9.02-01	-0.03	9.78-01
	Severe hADRs	-3.65	1.06-03	-0.68	5.00-01
	Less severe hADRs	-2.74	9.73-03	-0.58	5.65-01
DrugPoints	Liver failure	-0.82	4.20-01	0.42	6.75-01
	Jaundice	-0.11	9.15-01	1.18	2.47-01
	All hADRs	-0.81	4.21-01	0.04	9.67-01
	Severe hADRs	-1.37	1.78-01	-0.03	9.74-01
	Less severe hADRs	0.85	4.01-01	-0.41	6.81-01
DailyMed	All hADRs	0.00	1.00+00	0.00	1.00+00
	Severe hADRs	5.22	6.75-06	-0.60	5.50-01
	Less severe hADRs	1.41	1.72-01	10.04	1.57-10

For each endpoint, the AUC score vectors of model performance on all features were paired up and compared. ; .