QSPR APPROACH FOR MELTING POINT PREDICTION

摘要

Knowledge of physical properties and thermodynamic data are basic requirements for all Computer Aided Molecular Design applications. Basic properties, such as boiling and melting points, are important for developing specialty chemicals such as alternative refrigerants and extractive solvents. Currently, accurate correlations for melting point are limited, and recent attempts to use Quantitative Structure Property Relationship (QSPR) modeling have been inadequate. This is because (a) melting point predictions are highly sensitive to settle variations in molecular structure, and (b) available molecular descriptors do not satisfactorily address structure-property relationships. Specifically, the approach typically taken in constructing QSPR models is to select available, in-built molecular descriptors capable of representing the desired behavior. Although this approach has been successful in developing models for several physical properties, it fails for melting point prediction. In this case, the descriptors needed to predict the melting point of a molecule change as the subset is augmented thereby reducing the prediction efficiency of the model. In this work, we have developed a novel QSPR technique combining genetic algorithms (GA) and heuristic selection for choosing a set of relevant descriptors and building an optimal QSPR model for melting point prediction. This technique differs from other similar variable-selection techniques in that (a) the descriptor selection is performed by a genetic algorithm, and (b) the search is constrained to a maximum of ten descriptors and five operators. The results indicate that our 19-parameter model, which incorporates three newly constructed descriptors, is capable of modeling melting point behavior of 1250 organic chemicals with an average absolute percent deviation (%AAD) of 4.7% and R~2 value of 0.95. These results suggest that molecular weight and molecular connectivity play important roles in determining the melting point of a molecule.