Computational challenges and opportunities in spatially resolved transcriptomic data analysis

Abstract

Spatially resolved transcriptomic data demand new computational analysis methods to derive biological insights. Here, we comment on these associated computational challenges as well as highlight the opportunities for standardized benchmarking metrics and data-sharing infrastructure in spurring innovation moving forward.

Belayer: Modeling discrete and continuous spatial variation in gene expression from spatially resolved transcriptomics

“…We describe these ranking approaches below. Ranking genes by their Moran’s I score, a standard measure of spatial autocorrelation [67] Ranking genes by their Geary’s C score, another standard measure of spatial autocorrelation [35] Ranking genes by their likelihood ratio test (LRT) p-value, where we compare the maximum loglikelihood (12) with a piecewise constant expression function f g to the maximum log-likelihood (12) with a piecewise linear expression function f g . Ranking genes by their LRT p-value as above but restricted to Belayer Layer 3 (Figure 3D) Ranking genes by the sum of their (absolute) layer-specific slopes across the L layers identified by Belayer. Ranking genes by the maximum difference between layer-specific expression functions ) at layer boundaries Γ ℓ across the L layers identified by Belayer. …”

“…Ranking genes by their likelihood ratio test (LRT) p-value, where we compare the maximum loglikelihood (12) with a piecewise constant expression function f g to the maximum log-likelihood (12) with a piecewise linear expression function f g .…”

“…• Ranking genes by their Moran's I score, a standard measure of spatial autocorrelation [67] • Ranking genes by their Geary's C score, another standard measure of spatial autocorrelation [35] • Ranking genes by their likelihood ratio test (LRT) p-value, where we compare the maximum loglikelihood ( 12) with a piecewise constant expression function f g to the maximum log-likelihood (12) with a piecewise linear expression function f g .…”

“…Spatially resolved transcriptomics (SRT) technologies simultaneously measure both gene expression and spatial location of cells in a two-dimensional tissue slice [12, 72, 75]. Examples of SRT technologies include in-situ hybridization (ISH) techniques such as MERFISH [22] and seqFISH+ [33], which are based on imaging fluorescence probes, and sequencing techniques such as Slide-seq [76], Slide-seqV2 [80] and the 10X Genomics Visium spatial transcriptomics platform [1, 79], which sequence barcoded mRNA molecules whose barcodes record both the spatial locations of the molecules and unique molecular identifiers (UMI) for the mRNA molecule.…”

Scalable and model-free detection of spatial patterns and colocalization

“…Imaging-based approaches generally target pre-selected RNA or proteins at molecular and single cell resolution, while sequencing-based approaches allow genome-wide profiling with limited spatial resolution (Lewis et al 2021; Zhuang 2021). Recent advances in those approaches move the field rapidly into the direction achieving both high throughput and spatial resolution, presenting a significant computational challenge for scalable and robust methods to derive biological insights in the spatial context (Atta and Fan 2021).…”