SAR and QSAR modeling of a large collection of LD50 rat acute oral toxicity data

Table 2 Summary of modeling methods used

Method	Software	Descriptors	Applicability domain	Endpoints
Method	Software	Descriptors	Applicability domain	LD50 point estimate	vT	nT	EPA	GHS
BRF/rRF	KNIME	Dragon	“Error” model Confidence Similarity	✓	✓	✓	✓	✓
aiQSAR	R	Dragon	ADM	✓	✓	✓	✓	✓
istKNN	istKNN	Fingerprints + structural keys	Similarity/activity-based thresholds	✓
SARpy	SARpy	SAs	Presence/absence of SAs		✓	✓
HPT-RF	R (Caret)	Dragon	Mirror matrix Isolation forest	✓			✓	✓
GLM	R (H₂O)	Dragon	NA			✓

For each method, the software, the descriptors used, the applicability domain definition and the modeled endpoints are specified. The methods listed are balanced random forest (BRF)/regression random forest (rRF); ab initio QSAR (aiQSAR); istKNN; hyper-parameter tuning random forest (HPT-RF); generalized linear model (GLM)

ISSN: 1758-2946