Abstract Precision oncology relies on the accurate characterization of spatial molecular distributions in cancer tissues to uncover critical biomarkers and guide clinical decision-making. However, the high dimensionality and complexity of mass spectrometry imaging (MSI) data pose significant challenges for effective analysis. This study presents an unsupervised manifold learning framework to address these challenges by mapping high-dimensional MSI data into a low-dimensional space while preserving essential molecular patterns. This method enables efficient dimensionality reduction, clustering, and visualization of MSI data, facilitating the discovery of spatially resolved molecular features. Applied to datasets from prostate cancer and colorectal adenocarcinoma, the proposed method accurately identifies cancerous regions and reveals highly correlated molecular markers with Pearson correlation coefficients up to 0.79. These findings demonstrate the potential of unsupervised manifold learning to enhance the interpretability and utility of MSI data in precision oncology, paving the way for improved biomarker discovery and cancer diagnostics.
Jiang et al. (2026) studied this question.