Recent progress in multimodal artificial intelligence has shown that spatial transcriptomics data, which measure where genes are active within tissue, can potentially be generated from standard histology images, reducing the cost and time required for specialized experiments. However, newer spatial transcriptomics platforms such as Visium HD operate at very high resolution, down to about 8 micrometers, which creates major computational challenges. At this scale, traditional methods that predict gene expression one spot at a time become slow, unstable, and poorly suited to the extreme sparsity of gene expression, where many genes have very low or zero signal. To address this, the authors developed Img2ST Net, a high-resolution framework that predicts spatial transcriptomics data from histology images using a fully convolutional neural network, meaning the model generates dense gene expression maps all at once rather than sequentially. The method represents high-resolution spatial transcriptomics as groups of small regions called super pixels and reframes the task as an image generation problem with hundreds or thousands of output channels, each corresponding to a gene. This approach improves efficiency and better preserves spatial structure in the tissue. To evaluate performance under sparse expression conditions, the authors also introduced SSIM ST, a structural similarity-based metric designed specifically for high-resolution spatial transcriptomics. Testing on public breast and colorectal cancer Visium HD datasets at 8 and 16 micrometer resolution showed that Img2ST Net outperformed existing methods in both prediction accuracy and spatial coherence, while reducing training time by up to 28 times compared with spot-based approaches. Additional analyses showed that contrastive learning further improved spatial fidelity. Overall, this work provides a scalable and biologically meaningful solution for predicting high-resolution spatial transcriptomics data and supports future large-scale and resolution-aware spatial omics modeling.

Fig.��1

Modeling paradigm for ST prediction. (a)��Conventional patch-to-spot regression manner for Visium ST data: each WSI contains hundreds of��55μm��spots for the ST slide. A separate gene expression vector is predicted for each spot from its corresponding image patch. (b)��Our proposed image-to-image prediction framework for Visium HD data: each WSI contains millions of��8μm��bins for the HD slide. A region-wise modeling strategy where each image region covers multiple bins is used to predict a high-resolution gene expression map, which enables more fine-grained and computationally efficient inference.

]]> /valiant/2026/01/28/microenvironment-aware-spatial-modeling-for-accurate-inference-of-cell-identity/ Wed, 28 Jan 2026 15:26:03 +0000 /valiant/?p=5652 Liu, Qi; Wang, Yu; Hsu, Chihyuan; Wanjalla, Celestine N.; Lau, Ken S.; & Shyr, Yu. (2026).��.��Nucleic Acids Research,��54(1).��

Spatial omics technologies make it possible to measure many molecular features in cells while also preserving information about where those cells are located in a tissue. This spatial context provides valuable insight into how cells are organized and how tissues are structured. New platforms that work at single-cell resolution have further improved our ability to detect cell states that depend on the surrounding microenvironment. However, most existing computational tools for analyzing spatial omics data focus on identifying broad spatial regions rather than determining the identities of individual cells.

Traditional single-cell clustering methods define cell identities using only molecular features inside the cell, such as gene expression, and do not account for how nearby cells and the local tissue environment influence cell behavior. To address this limitation, we introduce MEcell, a method that directly incorporates spatial information and automatically determines how much influence the surrounding environment should have when identifying cell types. MEcell does not require users to tune parameters, making it easier to apply across datasets.

We tested MEcell on 90 simulated datasets and 7 real-world datasets from multiple spatial transcriptomics platforms and tissue types, including MERFISH Vizgen, Xenium, CosMx, Visium HD, Slide seq V2, and open ST. Across all tests, MEcell consistently performed better than existing methods at accurately identifying cell identities. These results show that the local microenvironment plays a crucial role in defining cell identity and demonstrate that MEcell is a powerful tool for capturing the full diversity of cells in spatial omics data.

Figure 1.

The rationale of MEcell. (A) A toy example where a cell’s transcriptionally similar neighbors are located within the same microenvironment, indicating that the microenvironment will play a minimal role in shaping the nearest-neighbor graph. (B) A toy example where a cell’s transcriptionally similar neighbors are located across distinct microenvironments, suggesting that the microenvironment will play a significant influence in shaping the nearest-neighbor graph.

Most well-differentiated thyroid cancers (WDTC) respond well to treatment, but aggressive forms such as anaplastic thyroid carcinoma (ATC) are often deadly. To better understand how thyroid cancer develops and becomes more aggressive in both children and adults, we analyzed gene activity at the single-cell level in more than 423,000 cells collected from 81 tumor samples. We also used spatial transcriptomics to map where different tumor cells and surrounding support cells are located within 28 tumors, including rare tumors that contain both WDTC and ATC features, as well as pediatric diffuse sclerosing thyroid carcinomas. In addition, we examined gene expression patterns from five large thyroid cancer datasets to study supportive tissue, known as the tumor stroma.

Using these approaches, we identified a specific group of cancer-associated fibroblasts called POSTN-positive myofibroblasts (myCAFs) that are located very close to invading tumor cells. The presence of these cells is strongly linked to worse outcomes, including cancer spread to lymph nodes and overall disease progression. We also found a different type of fibroblast, called inflammatory CAFs, that are located farther from tumor cells and are more commonly found in inflamed thyroid tissue, such as in autoimmune thyroiditis. Together, these findings show how thyroid cancers evolve within their surrounding tissue and identify a fibroblast subtype that could help predict aggressive disease in both children and adults.

Figure 1

Integrated single-cell atlas of thyroid cancer progression.

(A) Oncoplot for thyroid cancer publicly available single-cell RNA-sequencing samples: Luo et al. (26), Pu et al. (27), Lu et al. (28), Han et al. (29), Hong et al. (30), Lee et al. (31), Wang et al. (32). (B) Uniform manifold approximation and projection (UMAP) plot depicting the single-cell atlas labeled by broad cell type. (C) Scaled dot plot showing canonical markers for broad populations from��B. (D) UMAP colored by tumor histology with broad groupings of ATC (blue), PTC (light orange), or paratumor/normal (Para, gray). (E) Bar plots showing overall broad cell type composition for each paper in the single-cell atlas split by tumor histology group. pDC, plasmacytoid dendritic cell; NK/T, natural killer/T cell.

Recent advances in spatial transcriptomics (ST), which maps gene activity across tissue, show the need to analyze multiple tissue slices together. A major challenge is creating meaningful representations (embeddings) that preserve the tissue’s spatial layout while correcting for differences between slices (batch effects). We introduce��MaskGraphene, a graph neural network that integrates ST data using masked self-supervised learning, triplet loss, and cluster-wise local alignment. By forming indirect “soft-links” and direct “hard-links” between slices, MaskGraphene produces joint embeddings that closely preserve spatial geometry. Compared to eight existing methods, MaskGraphene achieves better alignment and interpretability. It also improves downstream analyses, including identifying distinct tissue domains, reconstructing developmental trajectories, discovering biomarkers, and mapping brain layers, providing a robust tool for integrating ST data and extracting biological insights.

Fig 1

MaskGraphene workflow.��a��Illustration of spatial transcriptomics data integration scenarios addressed by MaskGraphene, including spatially consecutive slice pairs (I), simulated partially overlapping consecutive slice pairs (II), numerous spatially consecutive slices (III), temporally consecutive slices (IV), and horizontally consecutive slices (V).��b��Workflow of MaskGraphene: The preprocessing step organizes spatial coordinates and gene expression data. Inter-slice linkage is established through the construction of “hard-links” via cluster-wise local alignment and “soft-links” using contrastive learning with triplets. Embedding optimization leverages a masked graph autoencoder to generate batch-corrected joint embeddings by optimizing masked self-supervised loss () and triplet loss. Steps 1-5 of the main method are indicated with numbered circles.��c��Applications and evaluations (Step 6): (I) Interpretable joint embedding captures the original geometric structure. (II) Topographic map of brain slices with isodepth analysis reveals gene expression gradients across cortical layers. (III) Validation with simulated data demonstrates robust integration of partially overlapping slices. (IV) Trajectory inference reveals linearly connected developmental trends. (V) Alignment and integration of embryonic tissue structures enhance biomarker identification. (VI) Stitching of horizontally consecutive slices reconstructs spatially coherent regions

Metals such as iron and copper play important roles in cancer biology, but how they interact with genes, cell types, and tumor growth processes is still not well understood. In the past, metals and genes were usually studied separately, which made it hard to see how they influence each other. New technologies—like spatial transcriptomics for measuring gene activity in specific locations and elemental imaging for mapping metals across tissue—now make it possible to study these relationships simultaneously, though combining these data remains challenging.

In this proof-of-concept study, we examined metal-related signaling in the tumor microenvironment of a single colorectal cancer (CRC) tumor using a spatial multimodal workflow. This approach integrated elemental imaging, gene expression patterns, cell composition, and histopathology to identify metal-linked biological pathways. Our early results showed notable relationships—for example, high iron levels were found in areas with mesenchymal-like cells at the tumor’s growing edge, along with increased expression of genes involved in epithelial-to-mesenchymal transition and extracellular matrix remodeling. We also observed that regions with high copper concentrations overlapped with zones of active tumor growth and were associated with increased expression of immune response genes.

These findings, while based on a single sample, demonstrate that combining elemental imaging with spatial transcriptomics can successfully reveal how metals and gene activity interact within tumors. Applying this workflow to larger patient groups in the future may help uncover new biomarkers and therapeutic targets related to metal-driven tumor progression.

Figure 1.

Overview: spatial integration of spatial elemental imaging and spatial transcriptomics can reveal genes associated with metal bioaccumulation within specific tissue architectures, shedding light on metals-related pathways and cellular changes associated with tumorigenesis; BNEIR: biomedical national elemental imaging resource; TRACE: tissue region analysis through co-registration of elemental maps.

Spatial transcriptomics (ST) is a technique that links gene activity with the physical location of cells in tissue, providing detailed insights into how tissues function. Current methods either focus on cell location or tissue images, but none fully combine gene expression, histology (tissue structure), and precise spatial information in a single framework. These methods also often perform inconsistently across different datasets. To address this, we developed��stImage, an open-source R package that offers a flexible and comprehensive approach to ST analysis. stImage uses deep learning to extract features from tissue images and provides 54 strategies to integrate gene expression, tissue structure, and spatial data. We show that stImage works effectively across multiple datasets and helps users choose the best integration approach using a diagnostic graph. Overall, stImage improves the analysis of spatial transcriptomics data, enhancing our understanding of tissue organization. It is freely available at��.

Figure 1

Workflow of stImage. stImage comprises four main steps, image features extraction, preprocessing, data integration, and visualization. The different color squares with abbreviation of different modalities show what combination of modalities were processed for that strategy: G for gene expression; S for spatial coordinates; I for image features.

]]> /valiant/2025/08/25/powerest-statistical-power-estimation-for-spatial-transcriptomics-experiments-to-detect-differentially-expressed-genes-between-two-conditions/ Mon, 25 Aug 2025 20:28:54 +0000 /valiant/?p=5021 Shui, Lan, Maitra, Anirban, Yuan, Ying, Lau, Ken S., Kaur, Harsimran, Li, Liang, & Li, Ziyi. (2025). “.” PLOS Computational Biology, 21(7), e1013293.

Recent advances in��spatial transcriptomics (ST)—a technique that measures gene activity in tissue while preserving its location—have greatly improved biological research. However, current ST methods are expensive, making large-scale studies difficult. This creates a need to��make the most of available data��to achieve reliable results.

A key task in ST research is identifying��genes that behave differently under different conditions, known as��differentially expressed genes (DEGs). While these analyses are common, how to calculate their��statistical power—the ability to detect real differences—is rarely discussed.

To address this, we developed��PoweREST, a tool that estimates the power of DEG detection using��10X Genomics Visium data. PoweREST can be used��before starting experiments��or��after collecting preliminary data, making it flexible for many study designs. We also created a��user-friendly web applicationthat allows researchers to easily calculate and visualize the power of their ST studies without needing to write any code.

Fig 1. Schema of the proposed PoweREST method.

When a preliminary cohort of ST data is available, PoweREST performs the power calculation based on bootstrap and P-splines fitting. When preliminary data are not available, an R Shiny app with power estimation results based on datasets from two cancer studies can be used. Created in BioRender. Shui, L. (2025)