Benchmarking Spatial Clustering Methods for Mass Spectrometry-Based Spatial Metabolomics.

Area of Science:

• Spatial metabolomics
• Computational pathology
• Bioinformatics

Background:

• Mass spectrometry imaging (MSI) maps metabolite distributions in situ.
• Spatial clustering delineates metabolically distinct tissue regions.
• Systematic benchmarking of MSI spatial clustering methods is lacking.

Purpose of the Study:

• Evaluate ion filtering and clustering method selection effects on MSI spatial clustering performance.
• Establish a dual-metric framework for assessing spatial continuity and metabolic heterogeneity.
• Benchmark diverse clustering algorithms across varied MSI datasets.

Main Methods:

• Evaluated 30 clustering algorithms on 12 heterogeneous MSI datasets.
• Utilized a dual-metric framework assessing spatial continuity and metabolic heterogeneity.
• Included datasets from various ion sources, mass analyzers, and spatial resolutions.

Main Results:

• Noise filtering improved non-spatial methods' spatial continuity by ~28% but offered limited gains for spatially aware methods.
• Only 11 methods met both dual-metric criteria across datasets; SSC and DRSC performed well.
• Top-ranked methods showed 22% higher concordance with cell-type annotations than lowest-ranked methods.

Conclusions:

• The proposed framework and SMcluster platform offer standardized MSI clustering method benchmarking.
• Preprocessing and method selection are critical for spatial clustering performance in spatial metabolomics.
• Provides practical guidance for selecting and applying spatial clustering methods in MSI studies.