Identification Of Spatial Transcriptome Variable Genes

The development of spatial transcriptomics makes it possible to measure gene expression data and cells locations information simultaneously.Combined with cell location information,the amounts of gene expression at different locations are tested to identify spatial variable genes.Usually,cells only synthesize proteins required for their own structure and function,so the number of expressed genes and the amount of gene expression in each cell varies,which determine the cell’s shape and function.The identification of spatial variable genes can help to systematically analyze gene characteristics,cell state and biological function.Transcription factors is a kind of protein which can control chromatin and transcription and regulate the expression of specific genes by recognizing specific DNA sequences,so they have an impact on recognizing spatial variable genes.Previous studies on spatial variable genes did not consider the role of transcription factors in this process,and the analysis of the causes of spatial variable genes was not clear enough,which reduced the analysis efficiency.In this paper,the model TF-SVG is proposed to identify spatial variable genes and clarify the role of transcription factors.Firstly,the Poisson generalized linear model is established,and the transcription factors of the regulatory genes are selected by LASSO penalty.Secondly,the kernel-based independence test is constructed by using residuals and site coordinate data with the help of SPARKX method.Finally,Cauchy p-value combination rule is used to combine the p-values calculated under multiple kernel forms.In the empirical analysis of mouse brain spatial transcriptome data,the model improved the identification efficiency of spatial variable genes and found the source of spatial variability.The rationality of the model results was verified by analyzing the functions of typical genes and their related transcription factors and related diseases.