stars docs

SpaGRN

a comprehensive tool to infer TF-centred, spatial gene regulatory networks for the spatially resolved transcriptomic data.

SpaGRN is an open-source Python package for inferring gene regulatory networks (GRNs) based on spatial gene expression data using GPLv3 license. The model takes into account the spatial proximity of genes and TF binding to infer their regulatory relationships. The package is particularly useful for analyzing spatially resolved gene expression data. This approach can be applied to various types of spatial transcriptomics data, such as Stereo-seq, Seq-Scope, Pixel-seq, Slide-seq, MERFISH, STARmap, CosMx, ST, 10x Visium, DBiT-seq, and others. Notably, we are still working on the improvement of performance and calculation efficiency.

Highlights

Conceptual Advances:

(1) In the context of spatial transcriptomics (SRT) data, SpaGRN introduces a novel statistical approach leveraging spatial information to precisely identify cell type or functional domain-specific gene regulatory networks (GRNs) that determine the structure and function of localized cells. Instead of relying only on Pearson’s or Spearman’s correlation between gene expressions within the same cell, SpaGRN focuses on the spatially aggregated regulations, which theoretically outperforms existing scRNA-seq-based GRN inference tools, such as pySCENIC.

(2) SpaGRN employs motif enrichment analysis to transform bidirectional spatial co-expressions into directional causal relationships. In this way, SpaGRN reduces a number of false-positive transcription factor (TF)-target associations, as a common limitation in gene co-expression module inference methods, such as Hotspot, CellTrek, and SpaceX. These methods might incorrectly identify the co-expression of non-TF genes, which do not represent any regulatory relationships and should not be included in the inferred regulons.

(3) SpaGRN further incorporates spatial adjacency information to explore the influence of cell-cell interactions on the intracellular regulatory events through the identification of spatially co-expressed receptors and TFs, allowing for a more holistic view of the intercellular interactions to intracellular dynamics driving cellular regulations within the spatial context. This functionality sets SpaGRN apart from other approaches that solely focus on intracellular regulations, including all the competing tools mentioned above.

Methodological Innovations:

To fully exploit the information of spatial co-expression, regulatory direction, and receptor-involved influence, we have designed a series of algorithms to accommodate the information integration within the SpaGRN’s framework, and applicable to various types of input generated by different SRT platforms. Specifically, the original version of SpaGRN consists of four modules, including identifying genes with spatial patterns, determining pairwise spatial co-expression correlations for identified gene pairs, excluding false-positive correlations through motif enrichment analysis, and linking potential receptors to directional regulons. For instance, to capture the localized co-expressions and reduce the influence of high-order neighbors, SpaGRN incorporates a spatial kernel with decay weights in the first two modules.

To further improve the methodological capabilities and robustness, SpaGRN has integrated five spatial autocorrelation algorithms and three correlation algorithms to precisely detect both global gradients and local heterogeneity using ensemble learning in the revised version of SpaGRN. Notably, we have introduced the derivation of the bivariate Geary’s C method for the first time.

Workflow

Title figure