🧪 Technology & Methods Development
Impact and correction of segmentation errors in spatial transcriptomics.
Re-analysis of imaging-based spatial transcriptomics across multiple tissues and platforms shows that cell segmentation errors and molecule misassignment can severely bias core downstream analyses, including differential expression, neighborhood effects, and ligand–receptor interactions. The authors introduce a matrix factorization approach on local molecular neighborhoods that detects and mitigates these admixtures, analogous to doublet removal in scRNA-seq, substantially improving result reliability.
Impact: Establishes a practical framework to diagnose and correct segmentation artifacts, making spatial transcriptomics analyses more robust and interpretable.
Mitchel J et al., https://doi.org/10.1038/s41588-025-02497-4
Robust and interpretable prediction of gene markers and cell types from spatial transcriptomics data.
STimage is a deep learning suite that predicts spatial gene expression and classifies cell types directly from routine H&E images, while explicitly quantifying both data-driven and model uncertainty through ensemble foundation models. By integrating attribution maps, histopathology annotations, and functional gene information, STimage delivers interpretable predictions that generalize across platforms and can stratify patient survival and drug response.
Impact: Brings explainable, uncertainty-aware AI to digital pathology by inferring spatial molecular profiles from standard histology slides.
Tan X et al., https://doi.org/10.1038/s41467-026-68487-0
Integration of imaging-based and sequencing-based spatial omics mapping on the same tissue section via DBiTplus.
DBiTplus combines deterministic barcoding with RNase H-mediated cDNA retrieval to enable sequencing-based spatial transcriptomics and multiplexed protein imaging on the exact same tissue section. Applied to embryos, lymph nodes, and lymphoma FFPE samples, the method supports imaging-guided deconvolution to build single-cell–resolved spatial atlases and uncover mechanisms of lymphomagenesis and tissue organization.
Impact: Provides a unified multimodal workflow to jointly map transcriptomes and proteomes in situ, including in challenging clinical specimens.
Enninful A et al., https://doi.org/10.1038/s41592-025-02948-0
DPM: A Deep Learning and Optimal Transport Framework for Cost-Effective Spatial Metabolomics.
DeepPathMetabol (DPM) uses optimal transport–based deep learning to predict metabolite distributions in one MSI section from an adjacent section, achieving superior alignment and reconstruction over conventional similarity metrics. The framework can enhance MSI spatial resolution and reduce acquisition costs, and the authors highlight its conceptual extension to spatial transcriptomics and broader spatial biology.
Impact: Introduces an open-source optimal-transport framework that makes high-resolution spatial omics more affordable and scalable.
Yao B et al., https://doi.org/10.1021/acs.analchem.5c06903
SpatialFusion: A Unified Model for Integrating Spatial Transcriptomics to Unveil Cell-type Distribution, Interaction, and Functional Heterogeneity in Tissue Microenvironments.
SpatialFusion is a deep learning framework that integrates gene expression with spatial coordinates using graph neural networks, attention, and contrastive learning to jointly perform spatial domain detection and cell type deconvolution. Across benchmark datasets and breast cancer samples, it outperforms existing tools, better resolving fine tissue layers, capturing spatial heterogeneity, and nominating potential therapeutic targets such as COX6C and CCND1.
Impact: Delivers a powerful, unified model for high-resolution spatial domain mapping and deconvolution to dissect complex tissue microenvironments.
Wang M et al., https://doi.org/10.1016/j.jmb.2025.169535
🧠Neurobiology & Neurological Disease
Integrated Single-Nucleus and Spatial Transcriptomics Elucidate Heterogeneity and Hypoxia-Driven Organization of Supratentorial Ependymoma.
By combining single-nucleus and spatial transcriptomics in supratentorial ependymoma, this study charts the tumor’s cellular and spatial architecture, revealing two novel transcriptional programs linked to extracellular matrix remodeling and ZFTA fusion identity. Spatial scoring of hypoxia and mixed spots uncovers hypoxia-driven organization into distinct zones with differential immune infiltration, particularly a highly immune-enriched hypoxic niche.
Impact: Provides a spatially resolved blueprint of ependymoma heterogeneity that links hypoxia, transcriptional programs, and immune contexture, informing targeted therapy design.
SchĂĽftan E et al., https://doi.org/10.1158/0008-5472.CAN-25-2694
Spatially resolved single-cell analysis of transcriptomic changes linked with neuropathic pain in human neuromas.
Using single-nucleus RNA-seq and spatial transcriptomics on human trigeminal nerves and neuromas, the authors map cell types and transcriptional states associated with neuropathic pain. Painful neuromas show expansion of pro-inflammatory endothelial cells, upregulation of HLA-A, CXCL2, and CXCL8, and altered cell–cell signaling, with HLA-A protein expression correlating with pain severity.
Impact: Delivers a human, spatially resolved reference for peripheral nerve injury that pinpoints inflammatory and endothelial pathways as key drivers of neuropathic pain.
Morchio M et al., https://doi.org/10.1097/j.pain.0000000000003907
Integration of spatial and single-cell transcriptomic analysis uncovers cellular and molecular alterations in the hypertensive brain.
This study integrates spatial and single-cell transcriptomics across central nervous system regions involved in blood pressure control to characterize how hypertension reshapes cellular composition and gene expression. The analysis identifies region-specific molecular alterations and cell-type changes that may underlie dysregulated neurogenic control of blood pressure.
Impact: Provides a spatially anchored cell atlas of hypertensive brain regions, revealing candidate circuits and pathways for targeting neurogenic hypertension.
Gao Q et al., https://doi.org/10.1016/j.lfs.2025.124107
🩸 Hematology & Rare Disease
Single-Cell Profiling of ANKRD26 Thrombocytopenia Reveals Progenitor Expansion and Polyploid Apoptosis via JUNB-p21.
By integrating single-cell transcriptomics, spatial transcriptomics, and functional assays of bone marrow from multiple ANKRD26-related thrombocytopenia patients, this work reveals a conserved expansion of megakaryocyte progenitors and loss of polyploid megakaryocytes. Spatial localization and imaging show ANKRD26 enrichment at the centrosome, and mechanistic studies demonstrate that elevated ANKRD26 triggers apoptosis of polyploid megakaryocytes via a JUNB–p21 pathway independent of p53-PIDDosome signaling.
Impact: Offers the most detailed molecular and spatial characterization of ANKRD26 thrombocytopenia to date, uncovering actionable pathways to restore platelet production.
Chen L et al., https://doi.org/10.1182/blood.2025030017
🧬 Cancer Research & Tumor Microenvironment
NRF2-COX2-PGE2 axis drives immune cold tumors and predicts resistance to combination immunotherapy in hepatocellular carcinoma.
Focusing on hepatocellular carcinoma treated with atezolizumab plus bevacizumab, this study investigates tumor-intrinsic mechanisms that generate immune-cold microenvironments and therapy resistance. The authors identify an NRF2–COX2–PGE2 signaling axis as a key driver of immunosuppression, which correlates with poor response to combination immunotherapy.
Impact: Highlights the NRF2–COX2–PGE2 pathway as a biomarker and potential therapeutic target to overcome immune exclusion and resistance in HCC.
Yamamoto S et al., https://doi.org/10.1097/HEP.0000000000001677
Integration of imaging-based and sequencing-based spatial omics mapping on the same tissue section via DBiTplus.
DBiTplus is applied to human lymphoma tissues to jointly map spatial transcriptomes and protein expression, resolving cell types and signaling pathways that underlie lymphomagenesis, progression, and transformation. By preserving tissue architecture for multiplexed protein imaging after spatial barcoding and sequencing, the method connects transcriptional states with microenvironmental protein cues at near single-cell resolution.
Impact: Enables unprecedented multimodal dissection of lymphoma ecosystems on a single section, accelerating discovery of spatial biomarkers and therapeutic targets.
Enninful A et al., https://doi.org/10.1038/s41592-025-02948-0
SpatialFusion: A Unified Model for Integrating Spatial Transcriptomics to Unveil Cell-type Distribution, Interaction, and Functional Heterogeneity in Tissue Microenvironments.
In breast cancer tumor microenvironments, SpatialFusion leverages graph-based spatial learning to resolve fine-grained cell type distributions and identify spatially restricted functional programs. The model uncovers heterogeneous niches and proposes COX6C and CCND1 as spatially enriched candidate targets associated with tumor progression.
Impact: Offers a data-driven approach to decode spatially organized vulnerabilities in breast cancer, informing precision oncology strategies.
Wang M et al., https://doi.org/10.1016/j.jmb.2025.169535
🤖 Computational Spatial Omics & AI
Robust and interpretable prediction of gene markers and cell types from spatial transcriptomics data.
STimage ensembles foundation models to robustly predict spatial gene expression distributions and cell types from H&E images while quantifying both aleatoric and epistemic uncertainty. Integrated attribution and latent space analyses make the predictions biologically interpretable, enabling applications in prognosis and therapy response prediction.
Impact: Bridges imaging and omics with trustworthy AI tools that can be deployed in clinical workflows for molecular pathology.
Tan X et al., https://doi.org/10.1038/s41467-026-68487-0
DPM: A Deep Learning and Optimal Transport Framework for Cost-Effective Spatial Metabolomics.
Through optimal transport–guided deep learning, DPM aligns and predicts MSI-derived metabolite maps between adjacent sections, dramatically improving reconstruction fidelity. This strategy reduces the need for dense, high-cost imaging and conceptually extends to predicting other spatial omics layers.
Impact: Demonstrates how optimal transport and deep learning can upscale or impute spatial omics, cutting costs without sacrificing biological resolution.
Yao B et al., https://doi.org/10.1021/acs.analchem.5c06903
SpatialFusion: A Unified Model for Integrating Spatial Transcriptomics to Unveil Cell-type Distribution, Interaction, and Functional Heterogeneity in Tissue Microenvironments.
SpatialFusion’s dual encoding of spatial graphs and feature maps, coupled with self-supervised contrastive learning, enables accurate spatial domain delineation even in noisy, low-density datasets like human DLPFC. Its robust cell type deconvolution and interaction mapping generalize across tissues, providing a versatile backbone for spatial transcriptomics analysis.
Impact: Sets a new benchmark for integrated spatial domain identification and deconvolution, improving the analytical toolbox for spatial transcriptomics users.
Wang M et al., https://doi.org/10.1016/j.jmb.2025.169535