Boosting the analysis of IM-LC-MS data with Machine Learning

Accurate prediction of liquid chromatographic retention times and collisional cross section of metabolites is useful for better identification in untargeted MS and limiting experimental measurements in targeted MS. For example, retention time predictions have been applied to untargeted MS experiments to identify the correct lipids from multiple isobaric candidates. Even though proven useful, these predictions are not commonly applied in the analysis. This is mostly due to differences in experimental setups and as a result an incompatibility of models (differences in columns, gradients, solvents, stationary phase, etc.). We introduce the concept of generalized calibration. The key idea of this concept is to fit calibration curves on predictions from setup specific models. The calibrated predictions are then blended using machine learning for more accurate predictions on the experimental setup of interest.