Multi-view gene panel characterization for spatially resolved omics

Spatially resolved transcriptomics has transformed our ability to study complex tissues at the cellular and subcellular resolution. However, targeted spatial technologies require pre-selected gene panels, which are typically curated based on existing biological knowledge and prior research hypotheses. While current methods often prioritize capturing cell type information, we argue that an effective gene panel should also capture cell type diversity, cell states, pathway-level information, and minimize redundancy. To address these broader requirements, we developed a gene panel characterization platform that characterizes panels across multiple perspectives, thus allowing us to compare panels comprehensively. Notably, computationally constructed gene panels performed competitively in capturing major cell types when compared to our in-house manually curated panel. However, refined manual curation offered distinct advantages, particularly in capturing minor and rare cell types and exhibited lower information redundancy comparatively. Building on this framework, we integrated these metrics into a deep learning platform, panelScope, leveraging them as a loss function to design holistic gene panels. Using an acute myeloid leukemia (AML) dataset with 42 well-defined cell types and the 5K Xenium panel from 10X Genomics, we demonstrate the utility of our framework in comprehensively characterizing gene panels, enabling the design of tailored panels that address diverse research needs.