Understanding how gene regulatory programs reorganize across both space and time is essential for studying development and disease. Most existing methods infer regulatory structure from single timepoint spatial data, limiting their ability to resolve temporal remodeling of local gene–gene dependencies. The emergence of spatiotemporal transcriptomics datasets that measure spatial gene expression at multiple stages creates a need for methods that integrate information across time and space to infer time‑resolved spatial‑unit‑specific gene regulatory networks (GRNs).
The authors introduce SpaTemGRN to address this need. The framework is presented as a flexible, hypothesis‑generating approach that explicitly models observations from multi‑timepoint slides, with the specific aim of recovering how spatially localized regulatory relationships change across biological stages.
SpaTemGRN jointly models data collected from a sequence of spatial transcriptomics slides obtained at different timepoints or stages. A central feature of the approach is that it allows later‑stage spatial units to borrow statistical strength from units that are both spatially proximate and temporally preceding. This joint modeling contrasts with single‑snapshot strategies that treat each timepoint independently and therefore cannot share information across stages.
By integrating spatial proximity with temporal ordering, SpaTemGRN aims to improve the robustness and stability of inferred regulatory edges across stages. The developers position the method as particularly useful for detecting regulatory couplings that emerge, strengthen, or shift location over time, and for distinguishing broadly distributed early signals from later, spatially focal interactions.
The authors applied SpaTemGRN to spatiotemporal transcriptomics data from the App NL‑G‑F mouse model of Alzheimer’s disease. Using this dataset, SpaTemGRN revealed region‑ and age‑dependent strengthening of complement–glia regulatory coupling. Specifically, results indicated an early complement signature that was broadly distributed across spatial units at earlier stages, followed by later, spatially focal coupling between complement components and astrocytic and microglial responses.
These findings illustrate how the method can highlight a temporal sequence in which an initial, diffuse molecular signature precedes localized cellular regulatory interactions during disease progression. The study frames these observations as hypotheses about the spatiotemporal coordination of complement pathway activity and glial responses in this Alzheimer’s model.
SpaTemGRN was also applied to spatiotemporal data from the developing mouse embryonic brain. In this context, the method identified regulatory programs that became progressively sharper and more spatially segregated as the early neural tube regionalized. In other words, GRN structure inferred by SpaTemGRN suggested increasing spatial specification of regulatory relationships during embryonic regionalization.
This application demonstrates the framework’s utility for developmental biology, where spatial patterning and temporal progression are both critical to interpreting regulatory dynamics.
To assess performance, the authors compared SpaTemGRN to four existing methods across simulated datasets. In these simulations, SpaTemGRN recovered regulatory edges more robustly than the comparator methods and maintained the most stable performance across stages. These benchmarking results support the claim that joint modeling of multi‑timepoint spatial observations can enhance the reliability of inferred time‑resolved, spatial‑unit‑specific GRNs.
The source document reports these comparative advantages in general terms; detailed simulation parameters, evaluation metrics, and the identities of the four comparator methods are reported in the full preprint and supplementary material referenced by the authors.
All source code for SpaTemGRN is publicly released by the authors at https://github.com/yibingjiang/SpaTemGRN. The authors describe the framework as a hypothesis‑generating tool rather than a definitive causal proof of regulation; inferred edges represent putative regulatory relationships that merit further experimental validation.
The preprint format indicates that results have not yet undergone peer review. Where the source text omits implementation details, simulation specifics, or parameter settings, readers are referred to the preprint and its supplementary material for full methodological information.
This work is posted as a bioRxiv preprint (doi: https://doi.org/10.64898/2026.07.28.738702) and therefore has not been certified by peer review. The authors have declared no competing interests.