Prediction of methyl radical addition rate constants from molecular structure

Multiple linear regression and computational neural networks (CNNs) are use d to develop quantitative structure-property relationships for methyl radic al addition rate constants. Structure based descriptors are used to numeric ally encode substrate information for 191 compounds. Descriptors can be cla ssified as topological, geometric, electronic, or combination. A six-descri ptor CNN was developed that produced training set rms error = 0.381 log uni ts and rms error = 0.496 log units for an external prediction set. A seven- descriptor CNN was used to build a model for a subset of 172 of the compoun ds. Training set rms error was 0.424 log units and prediction set rms error reduced to 0.409 log units. Model predictions were on the order of experim ental error.