Scientists propose a new algorithm for the analysis of spatial cell type components

On August 7, the team of Weiwei Zhai / Liang Ma of the Institute of Zoology, Chinese Academy of Sciences, published a research paper entitled SONAR enables cell type deconvolution with spatially weighted Poisson-Gamma model for spatial transcriptomics in Nature Communications. A novel algorithmic model for resolving spatial cell type components based on spatial transcriptome data was proposed and named SONAR.

Organisms are made up of various types of cells. The structural functions and dynamic changes of biological systems depend to a large extent on the spatial distribution of cells, and revealing and using their laws is of great significance for exploring life phenomena and exploring the evolutionary mechanism of diseases. Currently, spatial transcriptomics provides a powerful means to measure gene expression in tissues while preserving spatial information, thus providing the possibility to analyze the spatial distribution and function of cells. Due to the limitations of spatial resolution, gene expression at each spatial site (spot) of such techniques often comes from the mixing of multiple cells. Therefore, an efficient deconvolution algorithm is required to efficiently obtain the spatial composition of cell types. At present, the published algorithms for deconvolution spatial transcriptome data either fail to fully consider the characteristics of high sparseness and high noise of the transcriptome, or fail to make full use of the similarity information of spatial neighbors in the process of deconvolution, and problems such as inference errors or unrobustness often occur in practical applications.

The SONAR algorithm proposed in this study is a probabilistic model based on a space-weighted regression framework, which uses the Poisson-Gamma distribution to model the raw count of spatial transcriptome data, and can comprehensively consider location-specific shift and overdispersion of expression count according to the characteristics of spatial transcriptome data (Figure 1). In order to prevent excessive use of spatial information in tissue regions with high heterogeneity (e.g., the spatial composition of cells may change drastically across structural boundaries or tumors), SONAR simultaneously introduces three modules (spatial kernel function, preclustering, elastic weighting) to screen and effectively use spatial information.

Figure 1. SONAR algorithm workflow diagram

In this study, SONAR’s superiority over other algorithms in cell component resolution accuracy is verified on a large number of simulation sets with different local features (e.g., dominant type cell abundance/number of types, etc.) and different global features (e.g., spatial distribution/regional transition patterns, etc.), as well as on real single-cell precision spatial transcriptome datasets (mouse brain/human heart dataset). Among the many methods, only SONAR was able to resolve the subtle distribution of cardiac crest cells (cNCCs) and Schwann progenitor cells (SPCs) in the outflow tract region of the heart (Figure 2).

Figure 2. SONAR significantly improves the structural analysis of the cerebral cortex and the stable identification of the subtle structure of the heart

The work applied SONAR to the highly heterogeneous human pancreatic ductal carcinoma (PDAC) and human hepatocellular carcinoma (HCC) data, characterizing the spatial distribution of region-specific cell types. In HCC data, SONAR finely reveals trends in colocalization of immune cells and fibroblasts in the tumor microenvironment on the tumor/normal tissue transition region (Figure 3).

Bu 3, Colocalization trend of immune cells and fibroblasts on the transition region of SONAR liver tumors with normal tissues

In summary, SONAR, a new algorithm for the analysis of spatial cell type components using spatial information is developed, which is designed and applied to simulation sets with different spatial patterns, and analyzed and explored on various real datasets. With the popularization of spatial transcriptome technology and the accumulation of cell atlas data, SONAR will provide assistance for the analysis of life processes and disease progression, and the accurate exploration of large-scale spatial transcriptome data. The research work is supported by the National Key Research and Development Program of China and the National Natural Science Foundation of China. (Source: Institute of Zoology, Chinese Academy of Sciences)

Related paper information:

Special statement: This article is reproduced only for the need to disseminate information, and does not mean to represent the views of this website or confirm the authenticity of its content; If other media, websites or individuals reprint and use from this website, they must retain the “source” indicated on this website and bear their own legal responsibilities such as copyright; If the author does not wish to be reprinted or contact the reprint fee, please contact us.

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button