Estimation of the melting points, enthalpies of melting, and entropies of melting of organic compounds

摘要

Values for the melting points, total enthalpies of melting, and total entropies of melting of organic compounds are required in order to accurately estimate properties such as aqueous solubility and vapor pressure. These measurements, therefore, are critical to predicting how a chemical will behave both in the body and in the environment. Despite their importance, relatively few methods are available for predicting these properties from chemical structure. One reason for the lack of available methods is that these properties are easily obtained experimentally. The major reason, however, is that the development of general models for their prediction is extraordinarily challenging. This study first develops a model for estimating the melting points of organic compounds. The model incorporates additive functional group descriptors as well as non-additive descriptors of molecular geometry. The model is trained on the melting points of nearly 3000 compounds, has an R² value of 0.873 and an average error of 29.8 Kelvin degrees. The melting point model is then used to estimate the total enthalpy of melting through the incorporation of an additional geometric descriptor of molecular eccentricity, ε. Eccentricity is a measure of the extent to which the structure of a molecule deviates from a sphere. The total enthalpy model is trained on data for 191 compounds. The model has an R² value of 0.910 and an average error of 3812 J/mol. The total entropy of melting is then estimated from the predicted enthalpy value by incorporating an additional parameter in the model, τ, as a measure of molecular flexibility. The total entropy model is trained on data for the same 191 compounds as the enthalpy model. It has an R² value of 0.928 and an average error of 9.8 J/Kmol. The total enthalpy and entropy models have average absolute errors that are similar to those obtained using existing techniques that are more complex. The melting point method, however, is significantly more accurate and widely applicable than the additive models that are currently available.