Elaboration of a strategy for in silico screening for reversed-phase liquid chromatography method development

Reversed-phase liquid chromatography is one of the most widely used analytical methods for the analysis of mixtures of chemical compounds. The development stage can be extensive due to the multitude of possible stationary and mobile phases and analytical parameters to set up. Thorough screening of the different possible combinations is time-consuming and costly, even when using a systematic approach such as experimental planning. The development of quantitative structure-retention relationship (QSRR) models can accelerate this screening phase for mixtures of known composition by allowing "in silico" screening of experimental conditions and selecting pre-optimal conditions.
An experimental dataset composed of retention times of with ninety-eight diversified molecules was generated in the laboratory on three different HPLC systems (Waters Alliance) with gradients from 0% to 95% methanol in 20 and 60 minutes at five different pH (2.7, 3.5, 5, 6.5 and 8). The different buffers are common volatile buffers.
First, the molecular descriptors describing the different properties of each analyte are calculated. Then, relationships between experimental retention times and molecular descriptors are derived by different QSRR machine learning (ML) models for each condition. Next, a response surface model (RSM) is trained for each compound using the predicted retention times of the ML models. The last step is the multiple criteria decision analysis (MCDA) using the desirability index for the selection of the pre-optimal conditions. Three criteria are optimized: separation of the different analytes, robustness to the analysis parameters and maximum retention time per condition.
The results of the application of this strategy demonstrate that the combination of QSRR, RSM and MCDA offers the possibility to assist usefully the experimental screening phase by computational methods when developing chromatographic techniques for known sets of molecules.