MSI-ATLAS: Mass spectrometry imaging and explainable machine learning uncover the brain's lipid landscapes
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Recent advances in computational methods for Mass Spectrometry Imaging (MSI) have opened new frameworks for understanding the molecular composition and functional organization of organs in health and disease. Here, we present the first detailed molecular map of the brain lipidome across 123 brain regions, using MSI combined with trustworthy visualizations and 191 polygon annotations. Leveraging recent advances in MSI visualization, such as MSI-VISUAL, we mapped the annotations to anatomically and molecularly defined brain regions directly from the MSI data alone, without relying on external imaging modalities. To address challenges posed by annotation ambiguities and the inherent complexity of MSI data, we developed the Computational Brain Lipid Atlas (CBLA) - a novel graph-based visual explainability method that provides virtual landscapes of molecular lipid compositions across brain regions and functional subdivisions. CBLA serves a broader role beyond annotation refinement: it utilizes dimensionality reduction and ensembles of supervised machine learning models to enable interpretable exploration of the molecular data, clarifies relationships between annotations, explains the behavior of trained machine learning models, and facilitates the generation of biological hypotheses about functional organization and molecular linkages. Using CBLA, we uncovered new patterns of lipid distribution, functional integration, anatomical connectivity, and location-specific disease-associated molecular signatures. Through novel computational decomposition of annotated regions into m/z features, we performed large-scale, high-resolution mapping of the brain lipidome, further enhanced by a novel method for resolving full-precision m/z values from binned MSI data. This mapping may be used as a starting point for further research on the brain lipidome. We demonstrate its usage for multiple biological insights, including the molecular composition of Aβ plaques, their spatial distribution, and their relationship with other brain regions.