Sphinx: An R Package for Adaptive Spatial Neighborhood Modeling and Functional Analysis in Spatial Proteomics

Spatial proteomics reveals tissue microenvironmental heterogeneity through protein localization, yet existing bioinformatics tools remain primarily optimized for transcriptomics and struggle with high-dimensional proteomics datasets.

To bridge this technological gap, researchers developed Sphinx, an R package providing a comprehensive workflow for spatial proteomics data preprocessing, cell annotation, adaptive spatial neighborhood modeling, and functional characterization. The package was introduced in a study published on November 11, 2026, in PLoS Computational Biology.

Traditional bulk proteomics measures large cell populations while discarding critical spatial context. Spatial proteomics preserves tissue architecture at single-cell resolution, earning recognition as a vital method for decoding biological complexity in health and disease.

Despite its promise, the field faces steep computational demands due to explosive data growth. Most current analysis platforms rely on fixed-scale or single-radius strategies, which often overlook noise and multi-scale variations inherent in complex proteomics signals.

Development on the project began with a manuscript received on June 24, 2026, before gaining acceptance on October 27, 2026. The resulting tool integrates data processing, cell type identification, spatial organization modeling, and functional interpretation into a single streamlined pipeline.

The software operates through four core modules implemented in an R environment. Data preprocessing manages quality control for spatial proteomics data, utilizing the Seurat package to create a structured workflow.

A unified data interface function, load_spatial_data(), supports input formats including Seurat RDS objects, CSV coordinate files, and Akoya system outputs. Quality control filters low-quality cells using a median absolute deviation criterion.

Once data passes quality control, researchers utilize an interactive manual labeling module based on protein expression. The spatial neighborhood modeling framework then automatically selects among Delaunay, k-nearest neighbor, radius, and window-based strategies.

The adaptive parameter selection framework guides analysis using spatial statistical metrics. This approach handles varying tissue densities and signal heterogeneity better than rigid, single-scale alternatives.

The software identifies cell types and spatial functional units in lung adenocarcinoma, colorectal cancer, and mouse spleen datasets. It successfully detects structures such as tertiary lymphoid structures within the tumor microenvironment.

The package also accommodates paired high-resolution spatial transcriptomics datasets. This cross-technology compatibility stems from its reliance on spatial coordinates and feature matrices.

Functional analysis links these identified spatial patterns to biological pathways using databases such as KEGG, GO, CORUM, and Reactome. This connects structural tissue organization directly to underlying molecular pathways.

The computational optimizations built into the package allow for faster processing of large-scale biological datasets. These efficiencies address the scalability bottlenecks that frequently hamper high-dimensional spatial analysis.

Primary datasets utilized to validate the framework remain accessible for download through links provided in the original study. Raw data for lung adenocarcinoma come from the Kero HGC repository, while colorectal cancer and ovarian cancer data are hosted by the Spatch PKU platform.

Additional validation used pancreatic cancer data sourced from eLife archives. Financial support for the development came from the Science and Technology Commission of Shanghai Municipality and the Shanghai Municipal Health Commission Collaborative Innovation Cluster Project.

The authors declared no competing interests associated with the research.