A Robust Kernel-Based Workflow for Niche Trajectory Analysis

Niche trajectory analysis is a promising framework for modeling spatially continuous variations of the tissue microenvironment. However, the existing approach is limited by its requirement of cell-type annotation as a necessary input, which can lead to unwanted technical variations. To overcome this limitation, a new kernel-based strategy is presented that models the structural composition of a niche as a continuous function in the gene expression space, thereby obviating the need for cell-type annotation. Further integration with cell-type deconvolution analysis extends its application to datasets with any spatial resolution. Applying this strategy to real datasets indicates enhanced performance in robustness and accuracy and provides new insights into injury or disease-associated tissue microenvironment changes. As such, a useful tool for spatial transcriptomics data analysis is provided.