Functional domain identification in spatial transcriptomics transforms spatial molecular measurements into mechanistic insights about tissue physiology and pathology. The authors frame the need for methods that translate noisy molecular readouts into clear, biologically meaningful domains, emphasizing that improved domain detection can advance understanding of complex tissue architectures and disease microenvironments.
The study highlights two main barriers in current approaches. First, gene expression measurements in spatial transcriptomics are inherently sparse and noisy, which complicates reliable domain assignment. Second, many existing graph‑based approaches are biased toward local neighborhood structure, limiting their ability to capture higher‑order connectivity or long‑range relationships in the tissue microenvironment. Together, these factors can reduce the accuracy and biological interpretability of domain identification.
To address these challenges, the authors propose a method called Biologically Interpretable multi‑modal Graph using Spatial Transcriptomics (BIGraph‑ST). BIGraph‑ST aims to integrate complementary data modalities—molecular pathway activity and histological image features—with spatial transcriptomic profiles. The principal objectives are to (1) represent modality‑specific similarity through affinity graphs, (2) propagate spatial topology to capture higher‑order tissue connectivity, and (3) provide biologically interpretable pathway‑level representations of the resulting domains.
BIGraph‑ST represents each data modality by constructing an affinity graph that captures similarity among spatial spots or regions according to that modality. The approach explicitly integrates pathway activity scores—derived from transcriptomic data—and features extracted from histological images alongside raw gene expression. By modelling each modality’s similarity structure separately and then integrating these structures, BIGraph‑ST seeks to preserve modality‑specific information while enabling cross‑modal consensus on spatial domains.
Beyond constructing affinity graphs for individual modalities, BIGraph‑ST propagates spatial topology across the integrated graph to capture higher‑order connectivity. This propagation step is designed to overcome the locality bias of conventional graph methods by allowing information to diffuse beyond immediate neighbors. In doing so, the method captures more global patterns of tissue organization that may reflect biologically relevant interactions or microenvironmental structure.
According to the authors, experimental results showed that BIGraph‑ST delivered robust performance and notable improvements on multiple gold‑standard benchmark datasets. The manuscript specifically notes stronger performance in cancer tissue datasets, where complex and heterogeneous microenvironments challenge standard domain‑calling methods. The paper positions these results as evidence that integrating pathway activity and histological features, combined with spatial propagation, can yield more accurate domain delineation in spatial transcriptomics data.
A key contribution emphasized by the authors is the ability of BIGraph‑ST to provide biologically interpretable representations at the pathway level. Rather than producing black‑box domain labels, the method links identified spatial domains to pathway activity patterns, enabling mechanistic interpretation of domain identities. This feature is intended to help researchers derive biological hypotheses about tissue function, cell–cell interactions, and pathological processes from spatial transcriptomics experiments.
The work is presented as a preprint and has not been certified by peer review. The authors state that the source code for BIGraph‑ST will be made publicly available upon acceptance. The manuscript declares no competing interests. As with all preprints, readers should interpret results with consideration for the absence of formal peer review and await code release for replication and further evaluation.
BIGraph‑ST introduces a multi‑modal graph integration framework for spatial transcriptomics that explicitly combines pathway activity, histological image features, and spatial propagation to detect tissue domains. The method is reported to improve domain identification on benchmark datasets, with particular gains in cancer tissues, and to offer pathway‑level interpretability that supports biological insight. The preprint positions BIGraph‑ST as a tool to enhance mechanistic understanding of complex tissue architecture, while noting that broader validation and code availability will follow upon peer review and acceptance.