Biology subjects

Dumoulin, B.

Publications and source records attributed to Dumoulin, B..

6 recordsLinked to original sources

bioRxiv2026

Multiscale confidence quantification for virtual spatial transcriptomics with UTOPIA

Virtual spatial transcriptomics (ST) methods predict gene expression or cell types from histology images, extending molecular readouts beyond the limited regions or samples directly measured by ST platforms. However, the statistical reliability of these predictions remains unclear. Here, we present UTOPIA, a model-agnostic framework for multiscale confidence quantification in virtual ST. UTOPIA assigns statistically calibrated confidence scores to predictions across spatial resolutions and biological granularities, ranging from single genes to metagenes and from specific cell types to broader cell classes. UTOPIA controls false discovery rates for detecting genes, metagenes, or cell types while accounting for local tissue context. We show that prediction confidence depends critically on both spatial resolution and biological granularity, with reliable inference often emerging only at coarser, biologically meaningful scales. Across multiple ST platforms and in both in-sample and out-of-sample settings, UTOPIA enhances interpretability, prevents false biological conclusions, and enables more trustworthy downstream analyses of virtual ST.

genomics↗

bioRxiv2026

Pixel2Gene enables histology-guided reconstruction and prediction of spatial gene expression

Advances in spatial transcriptomics (ST) have fundamentally transformed our understanding of tissue biology by enabling gene expression profiling within intact spatial contexts and uncovering tissue organization and microenvironmental interactions. However, current high-resolution ST platforms remain constrained by high costs, limited tissue coverage, and technical artifacts, often yielding noisy, sparse, and incomplete data that compromise analytical accuracy, biological interpretation, and clinical utility. To address these challenges, we introduce Pixel2Gene, a deep learning framework that integrates co-registered histology images with ST data to enable histology-guided reconstruction and prediction of spatial gene expression. Pixel2Gene enhances existing expression measurements by denoising low-confidence data and reconstructing coherent expression patterns, while also predicting gene expression in unmeasured tissue regions and new samples lacking direct transcriptomic profiling. We systematically evaluated Pixel2Gene across multiple high-resolution ST platforms, including Visium HD, Xenium, and CosMx, spanning diverse tissue types and disease contexts using downsampling simulations and cross-platform comparisons in clinical samples. Across all settings, Pixel2Gene consistently improved data consistency, mitigated dropout effects, restored biologically meaningful spatial structure, and enabled accurate downstream analyses. By leveraging the scalability and ubiquity of routine histology, Pixel2Gene facilitates comprehensive, cost-effective ST profiling at whole-tissue scale, supporting large cohort studies, translational research, and next-generation biomarker discovery.

genomics↗

bioRxiv2026

HistoSweep enables cellular-resolution tissue quality control for gigapixel images in digital pathology and spatial omics

High-resolution histology images are indispensable for pathology and increasingly serve as the structural backbone for spatial omics. Yet whole-slide images (WSIs) frequently contain artifacts, acellular voids, and background regions that, when included in computational workflows, introduce noise, degrade model accuracy, and compromise biological interpretation. Existing tools provide only coarse foreground-background separation, leaving a gap in fine-grained quality control (QC). Here we present HistoSweep, a scalable framework that generates morphology-aware tissue masks at cellular resolution. By integrating density filtering, texture descriptors, and adaptive thresholding, HistoSweep systematically removes non-informative tissue regions while preserving biologically meaningful microstructures. It processes billion-pixel WSIs in minutes on standard CPUs, requiring no GPU acceleration, and is deployable across research and clinical settings. Across 25 WSIs spanning distinct tissues, disease states, and spatial omics platforms, HistoSweep consistently outperformed existing methods. It enhanced visualization and segmentation, improved virtual cell type predictions, and safeguarded spatial transcriptomics integrity by detecting transcript leakage and transcript-histology misalignment. By enabling fine-grained, scalable QC, HistoSweep provides a foundational preprocessing step for reliable and reproducible digital pathology and spatial omics analyses.

bioinformatics↗

bioRxiv2025

Nephrobase Cell+: Multimodal Single-Cell Foundation Model for Decoding Kidney Biology

BackgroundLarge foundation models have revolutionized single-cell analysis, yet no kidney-specific model currently exists, and it remains unclear whether organ-focused models can outperform generalized models. The kidneys complex cellular architecture and dynamic microenvironments further complicate integration of large-scale single-cell and spatial omics data, where current frameworks trained on limited datasets struggle to correct batch effects, capture cross-modality variation, and generalize across species. MethodsWe developed Nephrobase Cell+, the first kidney-focused large foundation model, pretrained on ~100 billion tokens from ~39.5 million single-cell and single-nucleus profiles across 4,319 samples, four mammalian species (human, mouse, rat, pig), and multiple assay modalities (scRNA-seq, snRNA-seq, snATAC-seq, spatial transcriptomics). Nephrobase Cell+ uses a transformer-based encoder-decoder architecture with gene-token cross-attention and a mixture-of-experts module for scalable representation learning. ResultsNephrobase Cell+ sets a new benchmark for kidney single-cell analysis. It produces tightly clustered, biologically coherent embeddings in human and mouse kidneys, far surpassing previous foundation models such as Geneformer, scGPT, and UCE, as well as traditional methods such as PCA and autoencoders. It achieves the highest cluster concordance and batch-mixing scores, effectively removing donor/assay batch effects while preserving cell-type structure. Cross-species evaluation shows superior alignment of homologous cell types and >90% zero-shot annotation accuracy for major kidney lineages in both human and mouse. Even its 1B-parameter and 500M variants consistently outperform all existing models. ConclusionsWith organ-scale multimodal pretraining and a specialized transformer architecture, Nephrobase Cell+ delivers a unified, high-fidelity representation of kidney biology that is robust, cross-species transferable, and unmatched by current single-cell foundation models, offering a powerful resource for kidney genomics and disease research.

genetics↗

bioRxiv2025

Designing smart spatial omics experiments with S2Omics

Spatial omics technologies have transformed biomedical research by enabling high-resolution molecular profiling while preserving the native tissue architecture. These advances provide unprecedented insights into tissue structure and function. However, the high cost and time-intensive nature of spatial omics experiments necessitate careful experimental design, particularly in selecting regions of interest (ROIs) from large tissue sections. Currently, ROI selection is performed manually, which introduces subjectivity, inconsistency, and a lack of reproducibility. Previous studies have shown strong correlations between spatial molecular patterns and histological features, suggesting that readily available and cost-effective histology images can be leveraged to guide spatial omics experiments. Here, we present S2Omics, an end-to-end workflow that automatically selects ROIs from histology images with the goal of maximizing molecular information content in the ROIs. Through comprehensive evaluations across multiple spatial omics platforms and tissue types, we demonstrate that S2Omics enables systematic and reproducible ROI selection and enhances the robustness and impact of downstream biological discovery.

genomics↗

bioRxiv2025

Single-Cell Spatial Mapping of Human Kidney Development Reveals the Critical Role of the Local Microenvironment in Cell Fate Decisions

Cell-cell interactions play a pivotal role in organ development, yet these communications have previously been studied one interaction at a time in model organisms, leaving a gap in our understanding of the cellular interplay in human development. To address this, we investigated human kidney development using single-cell RNA sequencing and spatial transcriptomics, analyzing over 500,000 cells. By mapping gene expression and differentiation trajectories in histologic space, we define the spatial organization of kidney development. Our analysis revealed newfound plasticity, showing that nephron progenitor cells undergo an early fate decision between renal corpuscle and tubular lineages. However, this choice is later reversed with some mature tubule cells transitioning back to a renal corpuscle fate. Further, through a genome-wide, spatially-aware cell-cell interaction analysis, we identified specific ligands and neighboring cell signals that create biologically meaningful cellular neighborhoods and mediate cell fate choices, offering a blueprint to understand the coordination of human development at scale.

genomics↗