Predicting the Ionization Behavior of Drugs in Tissue in MALDI and MALDI-2 Mass Spectrometry Imaging Using Machine Learning

Matrix-assisted laser desorption/ionization mass spectrometry (MALDI-MS) and its most common application, MALD-MS imaging (MSI), are widely used techniques in the analysis of intact biomolecules. In the context of pharmaceutical research, MALDI-MSI is often used to investigate the distribution of drugs and their metabolites within tissue section. While postionization strategies such as MALDI-2 have helped to increase signal intensities, ion yields in MALDI(-2) analysis are notoriously hard to predict. In many cases, this can complicate the planning and execution of pharmaceutical studies with regard to the expected limits of detection. To mitigate these challenges, we present a first approach to utilizing machine learning (ML) for the prediction of ionization efficiency. For this, we use data from a previously published data set containing MALDI and MALDI-2 data in positive and negative ion modes of ca. 1200 drug-like compounds acquired under imaging like conditions. To identify the optimal mode of action, we tested six different ML models and utilized selected physicochemical properties and 2-dimensional structures, both available for all employed compounds, for teaching. Subsequent SHAP analysis confirmed the involvement of a large number of parameters in the prediction as opposed to a dominant role of the presence or absence of a limited number of functional groups. In this, our proof-of-concept study highlights the usefulness of the multifactorial nature of ML to predict ion yields in MALDI(-2)-MSI. Beyond pharmacological application, the approach could, in the future, also assist in predicting ionizability in general MALDI-/MALDI-2-MSI measurements.