Linear solvation energy relationships as classifiers in non-target analysis – A gas chromatographic approach

Linear solvation energy relationships (LSERs) are applied as classifiers to predict the logarithmic retention factors log k from the structures of candidate compounds in non-target analysis. By comparison of the predicted value with the experimentally determined log k, progressive exclusion of candidates is done. The approach is based on the determination of stationary phase parameters to describe ten different gas chromatographic columns under four isothermal conditions. To demonstrate retention prediction and the application of the classifier model, twelve compounds with the molecular formula C12H10O2 were selected, while experimental log k values were compared to the predicted values and exclusion of potential candidate compounds was performed. The analytical power of the approach was demonstrated on the basis of experimentally determined compound descriptors achieved from gas chromatographic measurements. The prediction got less accurate when calculated compound descriptors were employed. For the time being insufficient precision in estimating the descriptors limits the possibility to exclude candidate compounds in non-target analysis. It is expected that new approaches to estimate compound descriptors, will improve this situation. At present, the insufficient accuracy of descriptor estimates can be dealt with larger prognosis intervals. Furthermore, the combination of two stationary phases with corresponding retention prediction further advanced the exclusion of potential candidates. The most appropriate pair of stationary phases was selected by the application of four different orthogonal strategies. In addition, the classifier was applied for a validation set with different molecular composition, where column selection was considered on the basis of the differences in the compound descriptors of the corresponding candidate compounds.