A set of molecular descriptors whose length is independent of molecular size is developed for machine learning models that target thermodynamic and electronic properties of molecules. These features are evaluated by monitoring performance of kernel ridge regression models on well-studied data sets of small organic molecules. The features include connectivity counts, which require only the bonding pattern of the molecule, and encoded distances, which summarize distances between both bonded and non-bonded atoms and so require the full molecular geometry. In addition to having constant size, these features summarize information regarding the local environment of atoms and bonds, such that models can take advantage of similarities resulting from the presence of similar chemical fragments across molecules. Combining these two types of features leads to models whose performance is comparable to or better than the current state of the art. The features introduced here have the advantage of leading to models that may be trained on smaller molecules and then used successfully on larger molecules.
[1]
Andreas Ziehe,et al.
Learning Invariant Representations of Molecules for Atomization Energy Prediction
,
2012,
NIPS.
[2]
Roman Garnett,et al.
BASC: Applying Bayesian Optimization to the Search for Global Minima on Potential Energy Surfaces
,
2016,
ICML.
[3]
Beat Ernst,et al.
Drug discovery today.
,
2003,
Current topics in medicinal chemistry.
[4]
Stéphane Mallat,et al.
Quantum Energy Regression using Scattering Transforms
,
2015,
ArXiv.
[5]
Wolfram Koch,et al.
A Chemist's Guide to Density Functional Theory
,
2000
.
[6]
J. Zico Kolter,et al.
An SVD and Derivative Kernel Approach to Learning from Geometric Data
,
2015,
AAAI.