Statistical analysis of spatial expression patterns for spatially resolved transcriptomic studies

Abstract

Identifying genes that display spatial expression pattern in spatially resolved transcriptomic studies is an important first step towards characterizing the spatial transcriptomic landscape of complex tissues. Here, we developed a statistical method, SPARK, for identifying such spatially expressed genes in data generated from various spatially resolved transcriptomic techniques. SPARK directly models spatial count data through the generalized linear spatial models. It relies on newly developed statistical formulas for hypothesis testing, providing effective type I error control and yielding high statistical power. With a computationally efficient algorithm based on penalized quasi-likelihood, SPARK is also scalable to data sets with tens of thousands of genes measured on tens of thousands of samples. In four published spatially resolved transcriptomic data sets, we show that SPARK can be up to ten times more powerful than existing methods, revealing new biology in the data that otherwise cannot be revealed by existing approaches.

Spatial multi-omic map of human myocardial infarction

“…We next investigated whether our spatial transcriptomics datasets reflected known biological processes of human myocardial infarction. To this end, we identified spatially variable gene expression across samples with SPARK 13 and identified overrepresented biological processes using hypergeometric tests. This analysis revealed functional and organizational differences consistent with the underlying biological conditions (Fig.…”

Unraveling Heterogeneity in Transcriptome and Its Regulation Through Single-Cell Multi-Omics Technologies

“…For instance, Trendsceek and SpatialDE are developed to identify genes with spatial expression pattern by incorporating both datasets (Edsgärd et al, 2018;Svensson et al, 2018). The more recent computationally efficient method, Spatial PAttern Recognition via Kernels (SPARK), displays superior statistical power as compared to the previous two methods (Sun et al, 2020).…”

Non-parametric modelling of temporal and spatial counts data from RNA-seq experiments

“…We compare our findings in figure 6(a), where GPcounts identified 468 spatially variable (SV) genes in total, whilst SpatialDE identified 67, with 62 of them overlapping between the two methods. The SPARK and Trendsceek methods, which consider calibrated p-values under a permuted null, identified 772 and 0 SV genes respectively (Sun et al, 2020). In figure 6(b) we plot the GPcounts log likelihood ratio versus SpatialDE log likelihood ratio, showing in quadrants the genes that are identified as SV by each method.…”

“…to identify differentially expressed genes, model temporal changes across conditions, cluster genes or model branching dynamics (Stegle et al, 2010;Kalaitzis and Lawrence, 2011;Äijö et al, 2014;Yang et al, 2016;Ahmed et al, 2019;Hensman et al, 2013;McDowell et al, 2018;Boukouvalas et al, 2018). GPs have also been applied to spatial gene expression data as a method to discover spatially varying genes (Svensson et al, 2018;Sun et al, 2020). In many previous applications of GP regression to counts, the log-transformed counts data are modelled using a Gaussian noise assumption.…”

“…Trendsceek (Edsgärd et al, 2018) assesses spatial expression of each gene by modelling normalised counts data as a marked point process, where the points represent the spatial locations and marks represent the expression levels, with a permuted null used to assess significance. SPARK (Sun et al, 2020) is a recently introduced GP-based method that relies on a variety of spatial kernels and a penalized quasi-likelihood algorithm. SPARK models counts using a Poisson likelihood function and captures over-dispersion through a nugget (white noise) term in the underlying GP covariance function.…”