Key points are not available for this paper at this time.
The emergence and development of spatial transcriptomics have provided a novel perspective for deciphering the heterogeneity of gene expression in spatial contexts. Unlike highly variable genes that disregard spatial context, spatially variable genes exhibit distinct spatial expression patterns within tissues. Current methods for detecting spatially variable genes tend to only emphasize overall statistical significance or the independent effects of covariates, while overlooking the interaction effects of covariates within the cellular microenvironment. To address this challenge, COVARE method is proposed based on the framework of a linear mixed-effects model, which can incorporate multiple predefined biological information and consider potential interaction relationships among covariates, thereby enabling precise analysis of multifactorial coregulation in gene spatial expression which is unachievable with analysis based on the assumption of covariate independence in existing methods. On this basis, the unbiased estimation of gene expression spatial effects achieved by COVARE is more consistent with the actual characteristics of the cellular microenvironment under physiological and pathological contexts, enhancing the application value in scenarios such as deciphering complex disease mechanisms and identifying potential biomarkers. Through extensive simulations and analyses based on public datasets, we have validated that COVARE delivers highly accurate detection and robust analytical performance. Additionally, genes with specific spatial expression patterns detected by COVARE provide a robust and biologically meaningful foundation for various downstream analyses of spatial transcriptomics.
Shi et al. (Sat,) studied this question.