TissueTrek：一个用于探索癌症空间形态学-分子关系的交互式多模态网络平台

Background: The spatial organization of epithelial, stromal, and immune compartments plays a fundamental role in shaping tumor behavior, yet researchers and clinicians lack accessible tools to concurrently explore morphology, pathomic features, and spatial molecular measurements. Existing methods typically examine a single modality at a time, require specialized expertise, or offer limited interpretability. To address this gap, we developed TissueTrek, an interactive web-based platform that integrates H&E morphology, quantitative pathomics, spatial gene and protein expression, and explainable Machine Learning(ML) outputs into a unified environment that can be easily used by researchers, clinicians, trainees, and patients alike. Methods: The platform was built using a GeoMx DSP cohort of 57 FFPE triple-negative breast cancer (TNBC) tissue cores. Board-certified pathologists annotated tumor regions of interest (ROIs). Three ROI-matched modalities were extracted: 77 pathomic features from H&E images, 91 spatially profiled and biologically relevant genes, and 570 spatially profiled proteins. The results from several ML pipelines developed upstream were incorporated to aid interpretability. Several web-based technologies enabled dynamic linking of ROIs with their molecular, pathomic, and model-derived representations for real-time exploration. Results: TissueTrek provides the first ROI-resolved interface enabling simultaneous visualization of (i) spatially measured gene and protein expression, (ii) quantitative pathomic features, and (iii) interpretable deep-learning and pathomics-ML outputs within the same tissue context. Users can interactively conduct exploration of results through multiple features to decipher the biology underlying the tissue ROIs. The Model Output panel displays interpretable maps of these ROIs, enabling the study of spatial morphology-molecular relationships in TNBC that are otherwise difficult to discern. Conclusions: TissueTrek represents a first-of-its-kind, fully interpretable spatial pathology interface that unifies morphologic, molecular, and deep-learning-derived insights into the same ROI-resolved environment. Although demonstrated in TNBC, the framework is generalizable and is capable of seamlessly incorporating additional biomarkers, diseases, or spatial-omics technologies, ​​ making it widely accessible, from expert computational scientists to clinicians, trainees, and patients seeking to understand their tissue biology. By enabling broad, user-friendly access to multimodal spatial biology, the platform advances computational pathology toward more transparent, clinically translatable, and educationally usable digital tissue ecosystem. This platform will be updated with the data and results constantly to serve the community better.