Related Experiment Video
Updated: Aug 6, 2026
10:16
Mining Spatial Transcriptomics Datasets using DeepSpaceDB
Published on: September 5, 2025
Spatial Autocorrelation Aware Resampling Improves Cell-Cell Interaction Inference in Spatial Transcriptomics Data
Parth Khatri1,2, Michael A Newton2,3, Christina Kendziorski2
1McArdle Laboratory for Cancer Research, Department of Oncology, University of Wisconsin-Madison.
Biorxiv : the Preprint Server for Biology
|July 17, 2026
Summary
A new method called Spatial Omics Autocorrelation-Aware Resampling (SOAR) improves spatial transcriptomics analysis by accounting for spatial autocorrelation. This reduces false positives and reveals distinct immune and resistance signatures in immunotherapy patients.
Area of Science:
- Computational Biology
- Spatial Transcriptomics
- Statistical Genetics
Background:
- Spatial transcriptomics enables analysis of cell-cell interactions by considering spatial signaling constraints.
- Existing statistical methods for spatial data lack the ability to account for spatial autocorrelation, leading to potential inaccuracies.
Purpose of the Study:
- To introduce a novel statistical method, Spatial Omics Autocorrelation-Aware Resampling (SOAR), for testing gene-gene correlations in spatial omics data.
- To address the limitation of spatial autocorrelation in existing significance testing methods.
Main Methods:
- SOAR utilizes spatial map patterns to decompose and maintain gene-level spatial autocorrelation in resampled datasets for null distribution generation.
- The method randomizes associations between gene expression and autocorrelation patterns to construct resampled datasets for significance evaluation.
- Validation was performed using simulation studies and analysis of 10X Visium and CosMX SMI spatial transcriptomics datasets.
Main Results:
- SOAR demonstrated maintenance of gene-level spatial autocorrelation and a reduced false-positive rate compared to random permutations in simulations.
- Application to HNSCC patient data identified a T-cell recruitment signature in responders and an angiogenesis/proliferation signature in non-responders.
- Analysis of a larger CosMX SMI cohort confirmed these trends, highlighting immune cell interactions in response to immunotherapy.
Conclusions:
- SOAR offers a statistically calibrated framework for analyzing spatial correlation in omics data by incorporating spatial autocorrelation.
- The method enhances the reliability of identifying biologically relevant gene-gene interactions in spatial datasets.
- Future work aims to connect localized correlation patterns to biological pathways for biomarker and therapeutic target discovery.
