Optimizing chromatographic conditions to obtain a good separation for analytes in complex samples is one of the great challenges in chromatography. An adequate separation of such mixtures is crucial for everything from detecting and identify impurities and degradation products in drugs, to detecting contaminants in wastewater. The main aim of this project is to develop decision-algorithms (e.g. reinforcement learning algorithms) to automate the decisions that have to be made to obtain an adequate separation for a complex mixture. Therefore, in this study, Markov Decision Processes (MDPs) will be introduced to optimize scouting runs that have to be made prior to building retention models for further method development.
Specifically, the percentage of modifier will be studied as the chromatographic parameter to be altered to obtain adequate separations. Modeling the retention factor k as a function of the percentage of modifier provides the ability to predict k’s of the compounds in the studied sample for all different percentages of modifier, for both isocratic and gradient runs. By optimizing the scouting runs with an MDP to obtain these models, the number of required experiments will be reduced significantly. The MDP will be trained for nearly one hundred representative small molecules under a large set of isocratic and gradient conditions. From this, the MDP will learn to decide the next step to take after starting with just one single isocratic scouting run, with a positive response/reward if the observations result in a good predictive model (measured as a mean-squared distance between predicted and experimental k). In future experiments, this MDP will be applied to unknown samples and will continue to learn based on the experimental outcome following the decisions made by the MDP.
Original languageEnglish
Publication statusPublished - Jan 2020
Event16th International Symposium on Hyphenated Techniques in Chromatography and Separation Technology - Het Pand, Ghent, Belgium
Duration: 29 Jan 202031 Jan 2020


Conference16th International Symposium on Hyphenated Techniques in Chromatography and Separation Technology
Abbreviated titleHTC-16
Internet address

ID: 49171416