AI Insight
Researchers developed multiScaleAC, a computational method that analyzes cell-cell interactions in spatial transcriptomic data across multiple spatial scales rather than requiring a single fixed radius. The method extends Moran's I statistical measure using a Gaussian kernel with varying bandwidth and applies functional data analysis to compare interaction patterns between samples. Validation through simulations showed well-controlled error rates and high statistical power, while application to renal cell carcinoma data revealed elevated COL4A1-ITGAV interactions in stromal tissues exposed to immunotherapy.
Why it matters
This approach addresses a key methodological limitation in spatial transcriptomics by reducing selection bias in defining cellular neighborhoods. The framework could improve understanding of tumor immune microenvironments and help identify therapeutic targets by more accurately capturing how cells interact across different spatial distances.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Over the last decade, spatial transcriptomic technology has transformed our understanding of tissue architecture including cell-cell interactions within the tumor immune microenvironment. A specific use-case of increasing interest is leveraging the spatial statistical relationship of genes whose protein products are known to be involved in ligand-receptor interactions. One methodological limitation of this approach has been the requirement to choose one radius around a cell as a parameter that can come with selection biases. Rather, interactions between cells vary in strength across a range of spatial scales that single-radius choice may miss. To fill this gap we developed `multiScaleAC` to extended Moran’s I, a correlation measure that accounts for locations of values, by employing a Gaussian kernel applied to locations and varying the bandwidth parameter h. The resulting Moranis Ileft(hright) then can be compared between samples using functional data analysis. In the current study, we used simulations to show that our framework has well controlled Type I error due to the use of permutations for assessing significant interactions. We also demonstrate that `multiScaleAC` has high statistical power to identify a significant interaction when a true interaction is simulated (1.00 at bandwidths greater than 5) and increasing power as bandwidth increases when negative interaction is simulated. We found `multiScaleAC` largely captures similar significant ligand-receptor profiles in 8 Visium samples of colon tissue using the same bandwidth as `spatialDM` but without removing low-weight spots from the weight matrix (73.7% – 87.2%). Applying the `multiScaleAC` framework to our previous single-cell spatial transcriptomics data (COL4A1-ITGAV in the stromal compartment of clear cell renal cell carcinoma) followed by functional principal component analysis, we found functional principal component 1 to represent global interaction elevation/depression. Associating functional principal component 1 scores with immunotherapy exposure showed significantly higher scores in stromal tissues exposed to immunotherapy than those naive to immunotherapy, indicating an overall higher interaction of cell expressing COL4A1-ITGAV. These findings recapitulate our previous study while reducing bias in neighbor selections. We believe this is the first study to apply a functional extension of Moran’s I in combination with functional data analysis to understand cell-cell interaction over spatial scales.
Source: multiScaleAC: Cell-Cell interaction with Moran's I as a function of kernel bandwidth