Evaluation of sequential Bayesian‐based method development procedures for chromatographic problems involving one, two, and three analytes

Abstract

In this work, various sequential Bayesian‐based method development procedures associated with a search of isocratic chromatographic conditions ensuring baseline separation of one, two, and three analytes within the pre‐specified retention time window were evaluated. The accuracy and total analysis time of tested procedures were experimentally and theoretically verified and compared with the usual approaches utilizing one or two preliminary organic modifier gradients. The possession of strong and weakly informative priors about analytes was considered. The overall success rate of tested procedures was high with a small number of mistakes. It was shown that for simple problems considered here one is able to use a sequence of isocratic conditions to find the desired chromatogram. The use of one preliminary gradient was shown to be of similar accuracy and time requirements. Interestingly, the use of two preliminary gradients almost always required more time than procedures involving no or one preliminary gradient. Prior knowledge of analytes considerably shortened method development times for procedures involving no or one preliminary gradient. The sequential Bayesian‐based method development allows us to achieve a desired chromatographic separation by proper weighting of information available to the analyst at each stage of the method development process.

Comparison of Chromatographic Stationary Phases Using a Bayesian-Based Multilevel Model

“…Because the proposed model provides a priori information for a subsequent analysis, this prior information can be utilized to decrease the number of future experiments. We can also perform experiments sequentially and then refine the predicted results after every experiment . The uncertainty-quantifying model allows identification of chromatographic conditions for which the desired separation is plausible at any state of knowledge related to the problem; this is equivalent to having access to some preliminary data.…”

Application of Bayesian Multilevel Modeling in the Quantitative Structure–Retention Relationship Studies of Heterogeneous Compounds

“…However, even in this case, the retention time/retention factor can be predicted, and knowledge about the likely chromatogram can be deduced. Information about the analyte structure is helpful and should be incorporated into decision making if possible. , The expected retention times, given the various sources of information (population-level parameters, functional groups, molecular mass, and any measurements), can be easily visualized, giving analysts a tool to make rapid decisions with regard to further steps: (i) stop method development if the desired separation is improbable, (ii) do more experimentation if current information is uncertain, or (iii) claim that this method is sufficient (i.e., performing more experiments is unlikely to show a better separation). Uncertainty chromatograms can also be used to quantify the probability of successful separation or to calculate the expected utility of the next experiment.…”

“…To make decisions with limited data, the analyst must consider a plausible range of chromatograms expected, given the available knowledge (i.e., analyte structure and/or experimental data) to make rational decisions. Bayesian methods are particularly well suited to show all necessary input to make decisions under uncertainty. − …”

Assisted Active Learning for Model-Based Method Development in Liquid Chromatography

“…Incorporated in AL methods, this data could speed up the MD process significantly. The incorporation of structural information on compounds, on the other hand, is a well-established practice in quantitative structure-retention relations (QSRR), wherein mathematical models are developed that relate the chemical structure of compounds to their chromatographic behavior, specifically, their retention times. ,− Although important progress has been made, estimating retention times in LC solely based on structural information with sufficient accuracy to support MD is still a complex endeavor. ,, Wiczling et al developed a Bayesian approach that combines both compound structure information and preliminary experiments to estimate the parameters of a retention model. , Although the concept was quite powerful on relatively simple cases involving one to a maximum of three analytes in isocratic conditions, it has not yet been incorporated into a fully automated MD process that strives to find an optimal method for more complex samples in gradient conditions …”