Research On Data Fusion Algorithm Of Single-cell And Spatial Transcriptomics Based On Topic Model

Single cell RNA sequencing(sc RNA-seq)can reveal the transcriptome information of cell subpopulation in a given organ and identify the heterogeneity of cell populations in tissues.In the process of experiment,sc RNA-seq needs to hydrolyze the tissue to obtain single-cell suspension,which will lose the spatial location information of cells.Spatial Transcriptomics can use tissue section and molecular labeling technology to capture tissue transcripts in situ and obtain the spatial location information and gene expression information of capture location,but it is difficult to achieve the resolution of single cell level.Therefore,the joint analysis of sc RNA-seq and spatial transcriptomics data can explore the spatial location information of single cells,obtain high-resolution tissue spatial expression map,and deepen the understanding of the interaction of specific cell subsets in development,regeneration and disease.To restore the spatial position of single cells and improve the resolution of spatial transcriptome spectrum,based on the improved semi-supervised topic model Space AX?Cor Ex and optimal transport,we combines with single-cell transcriptome data and spatial transcriptomics data and finds the cell composition in each capture position in spatial transcriptomics data.The proposed model can more accurately reveal cell interaction and tissue understanding,and provide new ideas for studying the whole development process of tissue.The main contributions of this paper are as follows:First,we propose a deconvolution model Space AX?Cor Ex to integrate single-cell and spatial transcriptomic data based on the semi-supervised topic model Anchor?Cor Ex.It introduces labeled single-cell RNA-seq data to analyze spatial transcriptomics based on semi-supervised topic model and Optimal Transport theory,which can reduce the experimental error from different batches and experiment platforms;Second,using the simulation data to intuitively compare the difference between the real values and results of Space AX?Cor Ex.We also compare it with other spatial transcriptomic analysis models such as Seurat,Spatial Dwls,Spotlight,Stereoscope,and Stride.Space AX?Cor Ex achieved the fewest errors.Third,the different developmental stages of human embryonic heart and the comparison of the temporal and spatial changes of different cell type composition are studied.We achieved the spatial transcriptomic map with single cell resolution during embryonic heart development.To sum up,this study provides a new framework for the joint analysis of single-cell transcriptomic data and spatial transcriptomic data.It makes a meaningful exploration and attempt to obtain the cell composition of the spot in spatial transcriptomic data and improve the resolution of spatial transcriptome spectrum.The experimental results on both the simulation data and real embryonic heart data show that our model achieves the fewest root mean square error and Jensen–Shannon divergence,which provides a new and powerful tool for the future research.