Computational Mass Spectrometry Imaging in the Era of AI

Abstract

Mass spectrometry imaging (MSI) is a tool-of-choice for mapping and understanding the spatial organization of biomolecules, including small metabolites, lipids, peptides, and many others. As the MSI instrument and spatial biology inquiries evolve, researchers and practitioners are constrained by the inherent trade-offs in spatial resolution, chemical detail, and acquisition time. Here, we review how the rapidly growing interplay between MSI and machine learning/artificial intelligence (ML/AI)-powered computational approaches is addressing these issues. We begin by highlighting key steps in MSI experiments and summarizing major ML/AI paradigms in the context of MSI data, providing a foundation to review how ML/AI impact each step in the MSI workflow, starting with methods to accelerate data acquisition. We then discuss emerging applications of dimensionality reduction, segmentation, and various supervised/unsupervised learning approaches to extract useful chemical insights from high-dimensional MSI data. Approaches to leverage multimodal imaging to guide the acquisition process or provide a more informative integrated analysis are discussed. We conclude with a forward-looking discussion on the state of computation and MSI, spanning ML-enabled instrumentation, scaling measurements to 3D and large cohorts, and the integration of MSI with other spatial omic data.