|Structure driven prediction of retention :Improvement of accuracy
|Dr Roman Szucs
Abstract Information :
Identification of a suitable starting point for chromatographic method development (stationary phase and mobile phase) is essential to ensure maximum robustness and ruggedness of the final method. In routine practice, this is typically achieved by screening a fixed number of stationary phases which offer significant differences in selectivity in combination with multiple mobile phases. Although such an approach represents a significant improvement over random selection of stationary phases, there are certain limitations. The entire selectivity space is not sufficiently covered, as only 4-6 stationary phases are typically screened with up to 6 mobile phases. Screening and subsequent data processing is time consuming, costly and also generates large amount of waste in the form of toxic solvents as well as data.
A study on the use of analyte chemical structures to permit prediction of retention times and thereby to select optimal chromatographic conditions is reported. In this approach, relevant molecular descriptors (features), generated from molecular modelling, are selected utilizing various statistical techniques and evolutionary algorithms. Mathematical relationships are then developed between selected features and measured retention times for training sets of compounds. These were chosen taking into account structural similarity of the "new compound", retention time of which is to be determined.
In this contribution we focus on optimization of retention models, selection of relevant features and provide initial insights on the impact of structural similarity between the training and test sets on accuracy of retention time prediction. The required size of the training sets will also be discussed for various scenarios of similarities between training sets and test compounds. Finally, the above described approach is applied to alternative separation mechanisms such as RP-LC, HILIC and Ion Chromatography.