Skip to main content

Advertisement

Table 1 Radial atomic reactivity descriptors [45] for the HBA/HBD atoms used for machine learning and kernel functions in Gaussian Process Regression (GPR) as implemented in scikit-learn 0.19.1 [82]

From: Machine learning models for hydrogen bond donor and acceptor strengths using large and diverse training data generated by first-principles interaction free energies

Descriptor abbreviation Description (for details, see our previous publication [45])
Sorted-shell Charge shell descriptor with values sorted by Cahn-Ingold-Prelog rules
CS Charge shell descriptor with average charge per shell
CRDF Spatial charge radial distribution function
CACF Spatial charge autocorrelation function (split into positive and negative parts)
MS Mass shell; the elements are the sums of the masses of each shell
GACF Topological charge autocorrelation function
  1. The hyperparameters of the constant kernel (C) and the RBF, M, and RQ functions were optimized in their default ranges (10−2 to 102 for length scales, 10−3 to 103 for C), and the white kernel (W) was used with a noise value of 0.05
GPR kernel function Description
\(C*RBF + W\) RBF = radial basis function (Gaussian)
\(C*M + W\) M = Matérn kernel function (v scanned manually for values of 0.5, 1.5 and 2.5)
\(C*RQ + W\) RQ = rational quadratic function
  1. The hyperparameters of the constant kernel (C) and the RBF, M, and RQ functions were optimized in their default ranges (10−2 to 102 for length scales, 10−3 to 103 for C), and the white kernel (W) was used with a noise value of 0.05