A basic task of analyzing spatial transcriptomics data is to identify spatially variable (SV) genes: here defined as genes whose expression distributions display significant dependence on their spatial locations. Besides the statistical characteristics, recent transcriptome-wide studies indicated that SV genes could also demonstrate a strong conservation in their spatial patterns, such that many SV genes display similar dependencies on spatial locations, resulting in similar trends in spatial patterns of their expression values5. Furthermore, SV genes are often markers or essential regulators for tissue pattern formation and homeostasis, consequently, the expression patterns of SV genes generally align remarkably well with underlying tissue structures. Recently, three prominent methods based on marked point process (trendSceek)7, Gaussian process (spatialDE)8, or Generalized linear spatial model (SPARK)9?were developed to identify SV genes. Although these methods have been shown to identify SV genes successfully, these algorithms have computational efficiency of?O(n2) or?O(n3)7,8,9, limiting their utilities as spatial transcriptomics data grows into millions of data points and beyond.
來自：Identification of spatially variable genes with graph cuts
