The effect of noise on the predictive limit of QSAR models

Table 1 Datasets used in this work, with the number of molecules, endpoint, endpoint units, range, and reference for each

Dataset	Category	Entries^a	Endpoint	Range	Refs.
G298_atom	Quantum mechanical	131,082	ΔG^o_at (kcal mol⁻¹)	− 2417to − 288	[29, 30]
Alpha	Quantum mechanical	131,082	α (Bohr³)	9.0 to 27.8	[29, 30]
Lip	Physiochemical	4200	logD	− 1.5 to 4.5	[31]
Solv	Physiochemical	642	ΔG^o_hyd (kcal mol⁻¹)	− 25.5 to 3.4	[32]
BACE	Biochemical	1513	pIC₅₀	2.5 to 10.5	[33]
Tox_102^b	Toxicological in vitro	971	logAC₅₀	− 2.1 to 4.7	[28]
Tox_134^c	Toxicological in vitro	1347	logAC₅₀	− 4.0 to 2.8	[28]
LD₅₀	Toxicological in vivo	5003	logLD₅₀ (mg kg⁻¹)	− 1.9 to 4.8	[35]

^aOriginal size of the dataset. If datasets have more than 1000 molecules, they were randomly sampled down to a size of 1000 before modeling
^bIncludes data exclusively from the ATG-PPre-cis assay
^cIncludes data exclusively from the ATG-PPARg-trans assay

ISSN: 1758-2946