Trend AnalysisBiology & Life Sciences

Spatial Transcriptomics: Mapping Gene Expression Where It Happens

Single-cell RNA sequencing revolutionised biology by revealing the transcriptomic identity of individual cells, but it requires dissociating tissues into cell suspensions โ€” destroying the spatial cont...

By Sean K.S. Shin
This blog summarizes research trends based on published paper abstracts. Specific numbers or findings may contain inaccuracies. For scholarly rigor, always consult the original papers cited in each post.

The Question

Single-cell RNA sequencing revolutionised biology by revealing the transcriptomic identity of individual cells, but it requires dissociating tissues into cell suspensions โ€” destroying the spatial context that gives tissues their function. Spatial transcriptomics restores this lost dimension by measuring gene expression in situ. The field faces a resolution-coverage trade-off: imaging-based methods (MERFISH, seqFISH) achieve subcellular resolution but measure hundreds of genes, while sequencing-based methods (Visium, Slide-seq) capture the whole transcriptome but at multi-cell spot resolution. Can computational methods bridge this gap, and what biological insights become visible only when spatial context is preserved?

Landscape

Zhao et al. (2024), in a paper introduced a transformer-based model that achieves super-resolution in spatial transcriptomics by integrating histology images with spatial gene expression data. Their approach computationally deconvolves Visium spots (55 ยตm diameter, covering ~10 cells) into single-cell resolution predictions. The key insight: H&E-stained histology images contain morphological information that, when fused with transcriptomic data via cross-attention transformers, can predict gene expression at resolutions below the physical spot size.

Lin et al. (2025) constructed the first integrated spatial atlas of human bone tissue, combining Visium spatial transcriptomics with single-cell RNA-seq. Their atlas revealed previously uncharacterised spatial niches within bone โ€” including a spatial gradient originating from trabecular bone, demonstrating systematic variations in gene expression, cellular composition, and signalling pathway activity (including TGFฮฒ/p-SMAD2/3 enrichment in peri-trabecular regions) that was invisible to dissociated single-cell studies. This work exemplifies the biological yield when spatial and single-cell modalities converge on the same tissue.

At the processing pipeline level, CellBin (Li et al. 2024) addressed a practical bottleneck: converting raw spatial transcriptomics imaging data into accurate single-cell gene expression matrices. Their pipeline achieved high-accuracy cell segmentation on Stereo-seq data, demonstrating that data processing โ€” not just data generation โ€” remains a rate-limiting step.

Methods in Action

The computational challenge in spatial transcriptomics centres on three tasks:

  • Deconvolution: Assigning cell-type identities to multi-cell spots using reference single-cell atlases. Xue et al. (2024) used graph convolutional networks (GCNs) to fuse spot-level transcriptomics, spatial coordinates, and histology features, achieving more accurate deconvolution than methods using transcriptomics alone. The graph structure naturally encodes spatial adjacency, allowing the model to borrow statistical strength from neighbouring spots.
  • Super-resolution: Predicting gene expression at finer spatial granularity than the measurement platform provides. Zhao et al.'s transformer approach treats this as an image-to-image translation problem, where the "image" is a gene expression map.
  • Whole-organism mapping: Wan et al. (2024) pushed spatial transcriptomics to its most ambitious scale โ€” whole-embryo mapping at subcellular resolution from gastrulation through organogenesis. Using multiplexed FISH across entire zebrafish embryos, they tracked how spatiotemporal gene expression patterns coordinate organ formation, revealing signalling gradients that operate over hundreds of micrometres.
  • Key Claims & Evidence

    <
    ClaimEvidenceVerdict
    Transformers can achieve single-cell resolution from Visium dataCross-attention model integrating histology + expression outperforms existing deconvolution methods (Zhao et al. 2024)Promising; validation against ground-truth MERFISH data needed for definitive claims
    Bone tissue contains spatially organised maturation nichesIntegrated spatial + scRNA-seq atlas reveals osteoblast gradient (Lin et al. 2025)Supported; consistent with known bone biology but first molecular-resolution mapping
    Graph neural networks improve spatial deconvolutionGCN fusing spatial, expression, and histology features outperforms non-spatial baselines (Xue et al. 2024)Supported across tested benchmarks
    Cell segmentation is a bottleneck for high-resolution platformsCellBin pipeline required for accurate single-cell assignment in Stereo-seq data (Li et al. 2024)Confirmed; a recognized infrastructure challenge

    Open Questions

  • Benchmark standards: How should computational super-resolution methods be validated? The field lacks gold-standard datasets where ground-truth single-cell spatial expression is known at Visium-equivalent locations.
  • Cross-platform integration: Can atlases built from different spatial transcriptomics platforms (Visium, MERFISH, Stereo-seq) be meaningfully compared or merged?
  • Temporal dynamics: Most spatial transcriptomics captures a single timepoint. How can time-series spatial data โ€” like Wan et al.'s embryo atlas โ€” be extended to postnatal development and disease progression?
  • Clinical translation: Pathology relies on H&E morphology. Can spatial transcriptomics add molecular layers to clinical diagnostics without requiring prohibitive costs?
  • What This Means for Your Research

    Spatial transcriptomics is shifting from a technology demonstration phase to a biological discovery phase. The papers reviewed here show that the highest-impact work now combines spatial data with computational innovation to answer specific biological questions โ€” not merely to generate atlases. For researchers in other fields: if your tissue of interest has a spatial organisation hypothesis (tumour margins, developmental gradients, organ-specific niches), spatial transcriptomics is now accessible enough to test it. The computational tools are maturing rapidly, but careful attention to cell segmentation and spot deconvolution remains essential to avoid artefactual conclusions.

    Referenced Papers

    • [1] Zhao, C. et al. (2024). Innovative super-resolution in spatial transcriptomics: a transformer model exploiting histology images and spatial gene expression. Briefings in Bioinformatics, 25(2). DOI: 10.1093/bib/bbae052
    • [2] Lin, W. et al. (2025). Mapping the spatial atlas of the human bone tissue integrating spatial and single-cell transcriptomics. Nucleic Acids Research. DOI: 10.1093/nar/gkae1298
    • [3] Xue, S. et al. (2024). Inferring single-cell resolution spatial gene expression via fusing spot-based spatial transcriptomics, location, and histology using GCN. bioRxiv. DOI: 10.1101/2024.10.27.620535
    • [4] Li, M. et al. (2024). CellBin: a highly accurate single-cell gene expression processing pipeline for high-resolution spatial transcriptomics. bioRxiv. DOI: 10.1101/2023.02.28.530414
    • [5] Wan, Y. et al. (2024). Whole-embryo Spatial Transcriptomics at Subcellular Resolution from Gastrulation to Organogenesis. bioRxiv. DOI: 10.1101/2024.08.27.609868

    References (5)

    Zhao, C., Xu, Z., Wang, X., Tao, S., MacDonald, W. A., He, K., et al. (2024). Innovative super-resolution in spatial transcriptomics: a transformer model exploiting histology images and spatial gene expression. Briefings in Bioinformatics, 25(2).
    Lin, W., Li, Y., Qiu, C., Zou, B., Gong, Y., Zhang, X., et al. (2025). Mapping the spatial atlas of the human bone tissue integrating spatial and single-cell transcriptomics. Nucleic Acids Research, 53(2).
    Xue, S., Zhu, F., Chen, J., & Min, W. (2024). Inferring single-cell resolution spatial gene expression via fusing spot-based spatial transcriptomics, location and histology using GCN.
    Li, M., Liu, H., Kang, Q., Fang, S., Li, M., Zhang, J., et al. (2023). CellBin: a highly accurate single-cell gene expression processing pipeline for high-resolution spatial transcriptomics.
    Wan, Y., El Kholtei, J., Jenie, I., Colomer-Rosell, M., Liu, J., Acedo, J. N., et al. (2024). Whole-embryo Spatial Transcriptomics at Subcellular Resolution from Gastrulation to Organogenesis.

    Explore this topic deeper

    Search 290M+ papers, detect research gaps, and find what hasn't been studied yet.

    Click to remove unwanted keywords

    Search 8 keywords โ†’