一个用于跨平台空间组学分析和深度空间分析的统一框架

Background: Spatial multi-omics has advanced rapidly in resolution, multiplexing, 3D modeling, and multi-modality. Platforms rely on distinct chemistries and assay architectures, yielding heterogeneous outputs but complementary strengths. These spatially indexed readouts enable neighborhood/niche inference, cell-cell communication, and 3D reconstruction, but demand rigorous normalization, uncertainty modeling, and scalable computation. Despite progress with foundation models for imputation and integration, a domain-specific suite that unifies these models for spatial omics is lacking. Methods: We developed Spyrrow (Spatially Resolved Multi-omics Data Visualization and Analytic Framework), a Python toolkit for cross-platform analysis, visualization, multi-modality integration and enhancement, spatial registration, and 3D reconstruction (https://github.com/WangLab-ComputationalBiology/spyrrow). Spyrrow ingests heterogeneous outputs and standardizes them into a unified spatial data model. Modules span raw-data processing (cell segmentation to cell-level matrices; single-cell enhancement for Visium; Visium HD single-cell transformation using registered H&E), multimodal registration, and a foundation-model hub (UNI, KRONOS, LOKI) for panel expansion, imputation, and cross-modality retrieval. With harmonized data and z-axis registration, Spyrrow builds 3D models linking serial sections. On the unified object, we provide clustering, graph-based neighborhood retrieval and niche identification, region detection (tumor compartments, TLS), spatial label transfer, and a comprehensive cell-cell communication module. Visualization includes pseudo-H&E/pseudo-fluorescence, ligand-receptor quiver plots, and 3D voxel renderings. Results: Spyrrow has been applied across multiple spatial-omics datasets and projects. The visualization suite produces clear, interpretable displays of niches, regions, and interaction patterns. Across benchmarks, integration with state-of-the-art models improves spatially aware clustering, label transfer, registration, and communication inference. Region-detection performance was corroborated by experimental validation. Together these results demonstrate robust performance, scalability, and cross-platform integration that are difficult to achieve with existing tools. Conclusions: Spyrrow addresses a critical need in spatial multi-omics by providing a unified, adaptable, and cross-platform analytical framework. It facilitates both biological discovery and methodological innovation by supporting a wide range of spatial omics platforms. Importantly, it offers a scalable and interpretable approach to spatial data analysis-from broad tissue-level context to detailed subcellular features-empowering deeper insights into the tumor microenvironment and advancing immune-oncology research.