Prediction of Retention Time of Morphine and Its Derivatives Without Using Computer-Encoded Complex Descriptors

A novel approach is introduced for reliable prediction of the retention time of morphine and its derivatives, which have widespread use in medical and pharmaceutical applications. A core correlation is introduced based on the number of carbon and hydrogen atoms as well as the number of ester, noncyclic ether, and ketone functional groups from experimental data of 42 compounds. An improved correlation is developed to increase the reliability of the core correlation by inserting a correcting function. The reliability of two correlations is tested and compared with the best available complex method based on a backpropagation artificial neural network (BP-ANN) for further 15 compounds. The values of root mean squared error (RMSE) of core and improved correlations for the test set are 0.5681 and 0.4832 min, which are lower than those reported by BP-ANN (0.6052 min). Cross-validations of the improved correlation corresponding to the coefficients of determination for leave-one-out (Q(LOO)(2)) and the fivefold cross-validation (Q(5CV)(2)) are close to its the coefficient of determination (R-2 = 0.982), which confirms goodness-of-fit, goodness-of-prediction, accuracy, and precision of the novel model.

Automated summary

A novel approach is introduced to predict the retention time of morphine and its derivatives reliably, based on experimental data of 42 compounds. An improved correlation is developed to increase reliability, and the method's accuracy is confirmed through cross-validation. Comparisons with a complex BP-ANN method show lower root mean squared errors and good predictive performance.