The effect of noise on the predictive limit of QSAR models

Table 3 Slopes m_noise and m_true for each dataset and algorithm

Dataset	Slope	Ridge	kNN	SVR	RF
G298_atom	m_noise	0.98 ± 0.011	0.79 ± 0.032	0.71 ± 0.032	0.79 ± 0.030
G298_atom	m_true	0.090 ± 0.010	0.09 ± 0.011	0.08 ± 0.00	0.09 ± 0.013
Alpha	m_noise	0.79 ± 0.033	0.83 ± 0.037	0.87 ± 0.023	0.89 ± 0.032
Alpha	m_true	0.12 ± 0.016	0.10 ± 0.012	0.12 ± 0.014	0.11 ± 0.013
Lip	m_noise	0.40 ± 0.031	0.36 ± 0.024	0.44 ± 0.024	0.41 ± 0.031
Lip	m_true	0.020 ± 0.011	0.021 ± 0.010	0.062 ± 0.013	0.032 ± 0.012
Solv	m_noise	0.75 ± 0.031	0.81 ± 0.031	0.89 ± 0.033	0.72 ± 0.031
Solv	m_true	0.13 ± 0.022	0.27 ± 0.011	0.27 ± 0.012	0.12 ± 0.012
BACE	m_noise	0.52 ± 0.042	0.53 ± 0.041	0.67 ± 0.033	0.54 ± 0.031
BACE	m_true	0.041 ± 0.021	0.052 ± 0.011	0.23 ± 0.023	0.050 ± 0.011
Tox_102	m_noise	0.44 ± 0.031	0.49 ± 0.043	0.44 ± 0.031	0.43 ± 0.031
Tox_102	m_true	0.010 ± 0.00	0.053 ± 0.011	0.00*^a	0.010 ± 0.00
Tox_134	m_noise	0.52 ± 0.034	0.57 ± 0.034	0.55 ± 0.031	0.49 ± 0.033
Tox_134	m_true	0.01*^a	0.041 ± 0.00	0.010 ± 0.00	− 0.020 ± 0.010
LD₅₀	m_noise	0.44 ± 0.042	0.43 ± 0.042	0.48 ± 0.033	0.48 ± 0.031
LD₅₀	m_true	0.00 ± 0.010	0.044 ± 0.016	0.083 ± 0.012	0.033 ± 0.012

^aEntries marked with * have p-values above 0.05 and thus are not statistically significant

ISSN: 1758-2946