<p>Mass spectrometry imaging (MSI) is a label-free technique that enables mapping hundreds of molecules in biological samples with high sensitivity and molecular specificity. MSI experiments usually sample a virtual grid of pixels on a sample surface. A full mass spectrum is acquired at each pixel. Typically, a single MSI experiment generates hundreds of thousands of spectra, each containing thousands of molecular features. The size of MSI data keeps increasing with MSI technology improvements in spatial resolution and molecular coverage. Subsequent interpretation of the vast and complex MSI data is a major bottleneck for deriving biological conclusions from the experimental results. In chapter 1, I review recently emerging computational methods in MSI for data analysis and “smart” experiments. I also provide a outlook for a future paradigm shift in MSI with transformative computational methods.<br> </p> <p>In my research, I have developed several approaches for the analysis and mining of the MSI data in a data-driven manner. In chapter 2, I introduce a vendor-neutral data processing pipeline for visualizing ion images from MSI data, which supports both standard and unconventional MSI acquisition strategies. In chapter 3, a spatial segmentation method is described. This method combines matrix factorization and manifold learning to enable the identification of distinct cells or tissue subregions in an unsupervised manner. In chapter 4, I describe a self-supervised approach for identifying and clustering colocalized molecules using contrastive learning, which helps analyze molecular pathways in biological samples. In chapter 5, I introduce a precise image registration method for studying individual fibers in mouse muscle tissue using multimodal MSI and immunofluorescence imaging. The locations of different types of muscle fibers obtained from immunofluorescence images are registered to MSI space, which enables biomarker discovery based on spatially resolved metabolomics and lipidomics data.</p> <p><br> Computational methods also provide powerful strategies for enhancing MSI capabilities, such as throughput, molecular coverage, and specificity. In chapter 6, I describe a “smart” sampling method for enhancing the experimental throughput of MSI. In collaboration with Prof. Dong Hye Ye’s group at Marquette University, we have developed a deep learning algorithm for sparse sampling (DLADS), which dynamically estimates molecularly informative tissue locations and guides sampling in MSI experiments. We coupled DLADS with nanospray desorption electrospray ionization (nano DESI) MSI platform through software and hardware integration. This approach preferentially samples informative tissue locations and reconstructs high-fidelity\ ion images with sparse MSI data, which improves the throughput of nano-DESI MSI experiments by 2.3-fold.<br>  </p>