QSRR Study of GC Retention Indices of Essential-Oil Compounds by Multiple Linear Regression with a Genetic Algorithm

Quantitative structure–retention relationships (QSRR) for components of the essential oil of the plant Bidens pilosa Linn. var. Radiata were studied to enable prediction of their retention indices (I _R). A data set was selected consisting of the retention indices of 44 components of the essential oil with a range of more than 635 units. A suitable set of molecular descriptors was then calculated and the best-fitting descriptors were selected by using stepwise multiple linear regression (SW-MLR) and a genetic algorithm (GA-MLR) the selection of variables. Comparison of the results obtained indicated the superiority of the genetic algorithm over the stepwise multiple regression method for feature selection. The predictive quality of the QSRR models was tested for an external prediction set of nine compounds, randomly chosen from the 44 compounds. One GA-MLR model with five selected descriptors was obtained. This model, with high statistical significance (R ² _train = 0.977, SE (%) = 2.33, F = 243.275, R ² _pred = 0.978), could be used to predict the retention indices of the molecules with error <6%.