Categorization of 33 computational methods to detect spatially variable genes from spatially resolved transcriptomics data

Abstract

In the analysis of spatially resolved transcriptomics data, detecting spatially variable genes (SVGs) is crucial. Numerous computational methods exist, but varying SVG definitions and methodologies lead to incomparable results. We review 33 state-of-the-art methods, categorizing SVGs into three types: overall, cell-type-specific, and spatial-domain-marker SVGs. Our review explains the intuitions underlying these methods, summarizes their applications, and categorizes the hypothesis tests they use in the trade-off between generality and specificity for SVG detection. We discuss challenges in SVG detection and propose future directions for improvement. Our review offers insights for method developers and users, advocating for category-specific benchmarking.

Figure 1:. General analysis workflows of single-cell transcriptomic data and spatially resolved transcriptomic (SRT) data.

The left column shows a general analysis workflow for single-cell transcriptomic data with steps including highly variable gene (HVG) detection, cell clustering, and cluster-marker gene identification. The right column illustrates a workflow for analyzing SRT data with steps including spatially variable gene (SVG) detection, spatial domain identification, and domain-marker gene identification.

Figure 2:. Conceptual visualization of three SVG categories and timeline of 33 SVG detection methods.

a. Conceptual visualization of three SVG categories: overall SVGs, cell-type-specific SVGs, and spatial-domain-marker SVGs. The top row shows a tissue slice with two cell types and three spatial domains. From left to right, exemplar genes with colors representing the expression levels are shown for an overall SVG, a cell-type-specific SVG, and a spatial-domain-marker SVG, respectively. b. Publication timeline of 33 SVG detection methods. Colors represent three SVG categories: overall SVGs (green), cell-type-specific SVGs (red), and spatial-domain-marker SVGs (purple).

Figure 3:. A hierarchical summary of 33 SVG detection methods.

The hierarchical summary considers the methodological characteristics, including graph conversion, kernel-based patterns, availability of statistical inference, statistical inference types, and gene expression distributions. Colors represent three SVG categories: overall SVGs (green), cell-type-specific SVGs (red), and spatial-domain-marker SVGs (purple).

Figure 4:. Conceptual diagram for Section “Theoretical characterization of SVG detection methods that use frequentist hypothesis tests.”

This diagram illustrates the logical relationships among the three types of statistical tests (dependence test, regression fixed-effect test, and regression random-effect test) used by the 23 SVG detection methods that rely on frequentist hypothesis tests. The diagram also introduces the general form of statistical models upon which regression-based tests are performed and the corresponding null hypotheses for detecting SVGs.

Figure 5:. Synthetic spatial patterns oversimplify the spatial patterns observed in real SRT data.

a, Representative spatial patterns used in synthetic SRT data for evaluating SVG detection methods such as Trendsceek [37], SpatialDE [26], and others. b, Spatial patterns shown in the 10x Visium dataset [19] profiling a dorsal lateral prefrontal cortex sample. The color indicates the log transformed expression for genes SNAP25, MOBP, and PCP4.