. Author manuscript; available in PMC: 2023 Nov 1.

Published in final edited form as: Toxicol Appl Pharmacol. 2022 Sep 20;454:116250. doi: 10.1016/j.taap.2022.116250

Table 4.

Distribution of model-predicted DILI and DICT compounds in the Tox21 10K library with and without the application of the applicability domain (AD) using Random Forest (RF), Naïve Bayes (NB), eXtreme Gradient Boosting (XGBoost), Support Vector Machines (SVM), and a consensus of all four methods.

		RF	NB	XGBoost	SVM	Consensus
DILI	without AD	2,807 (35%)	3,745 (47%)	2,384 (30%)	2,065 (26%)	1,374 (17%)
DILI	with AD ^*	1,536 (19%)	1,838 (23%)	1,389 (17%)	1,268 (16%)	845 (11%)
DICT	without AD	4,420 (55%)	3,075 (38%)	2,532 (32%)	2,163 (27%)	1,469 (18%)
DICT	with AD ^*	2,357 (29%)	1,698 (21%)	1,502 (19%)	1,357 (17%)	953 (12%)

Tanimoto coefficient ≥ 0.4