DPM: A Deep Learning and Optimal Transport Framework for Cost-Effective Spatial Metabolomics

Mass spectrometry imaging (MSI) is a powerful technology in spatial metabolomics that enables the in situ detection and distribution analysis of metabolites in tissue sections. However, the high cost associated with high-resolution and multislice MSI acquisition remains a major limitation. Here, we introduce DeepPathMetabol (DPM), a deep learning-enhanced framework based on optimal transport theory, which accurately predicts spatial metabolite distributions in an MSI section using data from an adjacent section through an optimized mapping strategy. DPM achieves superior alignment and prediction accuracy, outperforming conventional feature similarity-based methods such as those using Euclidean or kernel-based metrics both with and without spatial distance weighting. We further demonstrated that the DPM framework can effectively enhance MSI resolution, providing a powerful tool for cost-effective and high-precision spatial metabolomics research. This approach also shows promising potential for extension to spatial transcriptomics. Collectively, our work establishes histology-facilitated MSI-to-MSI prediction as a versatile strategy for spatial biology research. DPM is open-source and available at https://github.com/LinShuhaiLAB/DeepPathMetabol.