Joint modeling of multi-timepoint spatial observations for time-resolved spatial-unit-specific gene regulatory network inference
TL;DR
SpaTemGRN provides a flexible hypothesis-generating framework for investigating how spatially localized gene–gene dependencies are remodeled across biological stages and recovered regulatory edges more robustly than four existing methods and maintained the most stable performance across stages.
Abstract
Background How gene regulatory programs reorganize across space and time is central to development and disease, but current methods infer regulatory structure from single snapshots. The emergence of spatiotemporal transcriptomics calls for methods that resolve regulation along both axes. Results We introduce SpaTemGRN, which jointly models observations from multi-timepoint slides in spatiotemporal transcriptomics to infer time-resolved spatial-unit-specific putative GRNs. Compared with existing methods, SpaTemGRN allows later-stage units to borrow statistical strength from spatially proximate and temporally preceding units. In the AppNL−G−F mouse model of Alzheimer’s disease, SpaTemGRN revealed region- and age-dependent strengthening of complement–glia regulatory coupling. An early, broadly distributed complement signature precedes later, spatially focal coupling with astrocytic and microglial responses. In the developing mouse embryonic brain, SpaTemGRN identified progressively sharpening and spatially segregated regulatory programs as the early neural tube regionalizes. Across simulated datasets, SpaTemGRN recovered regulatory edges more robustly than four existing methods and maintained the most stable performance across stages. Conclusions SpaTemGRN provides a flexible hypothesis-generating framework for investigating how spatially localized gene–gene dependencies are remodeled across biological stages. All source code for SpaTemGRN is available at https://github.com/yibingjiang/SpaTemGRN.