Yuanhua Huang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Yuanhua Huang has a total of 288 co-authors. Here are the top co-authors by number of shared publications:
- Joshua Wing Kei Ho
The University of Hong Kong
10 shared publications - Guido Sanguinetti
Theoretical and Scientific Data Science Group, SISSA, Trieste, Italy
6 shared publications - Ryohichi Sugimura
The University of Hong Kong, Hong Kong, China
6 shared publications - Yiming Chao
School of Biomedical Sciences, The University of Hong Kong, HKSAR
6 shared publications - Chen Qiao
School of Biomedical Sciences, University of Hong Kong, Pokfulam, Hong Kong SAR, China
5 shared publications - Oliver Stegle
Wellcome Sanger Institute, Cambridge, UK
5 shared publications
Latest Publications
- Open Access
BLOOM: Biologically Informed Flow Matching for Transcriptomic Virtual Embryo Modeling
Constructing a virtual embryo requires a generative representation of how heterogeneous cell populations change across developmental time. Here we introduce BLOOM, a biologically informed framework for conditional generation of developmental cell populations. BLOOM integrates temporal population context with developmental and regulatory priors through conditional flow matching to generate cell-expression distributions at queried stages in either temporal direction. Evaluation spans three mouse single-cell atlases covering prenatal development, visual-cortex maturation and multi-organ endothelial specialization. In comparisons with 22 time-series forecasting baselines, BLOOM ranks first for gene- and cell-expression correlations across all three datasets. Measured-stage evaluations support recovery of developmental expression patterns, population organization and marker-specific temporal relationships. Biological analyses connect these predictions to cell-state identities, gene programmes and regulatory associations. At unmeasured stages, interpolation and autoregressive history and future rollouts extend simulated expression dynamics between and beyond measured timepoints. Analyses of marker genes and temporal descriptors identify candidate differences in the timing of muscle-associated, neural and vascular-specialization programmes. Input-based gene perturbations further probe predicted responses of edited genes and associated molecular programmes across time. Together, BLOOM supports transcriptomic virtual embryo modeling by bridging temporal sampling gaps, exploring extended developmental trajectories and prioritizing gene-dependent hypotheses for experimental investigation.
Open AccessBLOOM: Biologically Informed Flow Matching for Transcriptomic Virtual Embryo Modeling
Constructing a virtual embryo requires a generative representation of how heterogeneous cell populations change across developmental time. Here we introduce BLOOM, a biologically informed framework for conditional generation of developmental cell populations. BLOOM integrates temporal population context with developmental and regulatory priors through conditional flow matching to generate cell-expression distributions at queried stages in either temporal direction. Evaluation spans three mouse single-cell atlases covering prenatal development, visual-cortex maturation and multi-organ endothelial specialization. In comparisons with 22 time-series forecasting baselines, BLOOM ranks first for gene- and cell-expression correlations across all three datasets. Measured-stage evaluations support recovery of developmental expression patterns, population organization and marker-specific temporal relationships. Biological analyses connect these predictions to cell-state identities, gene programmes and regulatory associations. At unmeasured stages, interpolation and autoregressive history and future rollouts extend simulated expression dynamics between and beyond measured timepoints. Analyses of marker genes and temporal descriptors identify candidate differences in the timing of muscle-associated, neural and vascular-specialization programmes. Input-based gene perturbations further probe predicted responses of edited genes and associated molecular programmes across time. Together, BLOOM supports transcriptomic virtual embryo modeling by bridging temporal sampling gaps, exploring extended developmental trajectories and prioritizing gene-dependent hypotheses for experimental investigation.
- Open Access
CLEAR-ST: Physics-informed probabilistic decontamination of spatial transcriptomics by modeling mRNA lateral diffusion
Spatial transcriptomics is a rapidly evolving technology that allows for the measurement of gene expression in a spatially resolved manner. However, one technical problem that occurs for many sequencing-based spatial transcriptomics platforms is the presence of mRNA lateral diffusion, where mRNA from one spot can bind to probes in another spot, leading to contamination and inaccurate gene expression measurements. In Visium-like assays, this artifact is often visible as structured out-of-tissue signal and boundary-associated expression halos, yet its magnitude, spatial decay, and directional bias vary substantially across samples. Here, we present CLEAR-ST, a physics-informed probabilistic framework for correcting diffusion-like contamination in spatial transcriptomics data. CLEAR-ST infers a latent clean expression field using a denoising autoencoder and links it to the observed counts through a graph-Laplacian forward contamination model with learnable diffusion parameters, finally evaluated with a selectable count likelihood. We first conducted a comprehensive comparison between 10X official and independently generated Visium samples, demonstrating that out-of-tissue count profiles are highly related to nearby in-tissue expression, more concentrated near tissue boundaries, and diffusion directions across genes are likely coherent. Across real samples with varying contamination burden, CLEAR-ST improved spatial domain recovery, increased gene-level spatial autocorrelation, and enhanced the biological specificity of downstream analyses such as marker gene discovery, pathway identification and cell type deconvolution. Compared to benchmark methods, CLEAR-ST showed consistent gains in clustering quality and concordance with manual annotations. Together, CLEAR-ST provides an interpretable and practical approach for diffusion-aware correction of capture-based spatial transcriptomics data.
Open AccessCLEAR-ST: Physics-informed probabilistic decontamination of spatial transcriptomics by modeling mRNA lateral diffusion
Spatial transcriptomics is a rapidly evolving technology that allows for the measurement of gene expression in a spatially resolved manner. However, one technical problem that occurs for many sequencing-based spatial transcriptomics platforms is the presence of mRNA lateral diffusion, where mRNA from one spot can bind to probes in another spot, leading to contamination and inaccurate gene expression measurements. In Visium-like assays, this artifact is often visible as structured out-of-tissue signal and boundary-associated expression halos, yet its magnitude, spatial decay, and directional bias vary substantially across samples. Here, we present CLEAR-ST, a physics-informed probabilistic framework for correcting diffusion-like contamination in spatial transcriptomics data. CLEAR-ST infers a latent clean expression field using a denoising autoencoder and links it to the observed counts through a graph-Laplacian forward contamination model with learnable diffusion parameters, finally evaluated with a selectable count likelihood. We first conducted a comprehensive comparison between 10X official and independently generated Visium samples, demonstrating that out-of-tissue count profiles are highly related to nearby in-tissue expression, more concentrated near tissue boundaries, and diffusion directions across genes are likely coherent. Across real samples with varying contamination burden, CLEAR-ST improved spatial domain recovery, increased gene-level spatial autocorrelation, and enhanced the biological specificity of downstream analyses such as marker gene discovery, pathway identification and cell type deconvolution. Compared to benchmark methods, CLEAR-ST showed consistent gains in clustering quality and concordance with manual annotations. Together, CLEAR-ST provides an interpretable and practical approach for diffusion-aware correction of capture-based spatial transcriptomics data.
- Open Access
stCNASim: Allele-aware spatial RNA-seq simulator enables systematic benchmarking of copy number inference
Spatial transcriptomics (ST) is revolutionizing the study of tumor evolution by enabling spatially resolved copy-number alteration (CNA) analysis. However, evaluating the accuracy and robustness of current single-cell (SC) and ST-specific CNA inference tools remains challenging due to the absence of ground-truth datasets. Here, we present stCNASim, an allele-aware spatial RNA-seq simulator that generates raw reads within realistic spatial contexts. We synthesized 46 benchmarking datasets across varying technical settings and spatial architectures to evaluate five widely used computational methods. Our analysis reveals that while SC-based methods adapt well to ST data, ST-specific methods successfully benefit from considering spatial autocorrelation but struggle under high spatial intermixing. The allele-aware methods CalicoST, Numbat, and XClone achieved top-tier performance with unique advantages in extreme scenarios, yet showed distinct sensitivities to low purity, mirrored alleles, and low coverage, respectively. By providing a scalable simulator and a rigorous benchmark, this work establishes a much-needed framework to guide and accelerate future tool development in spatial CNA analysis.
Open AccessstCNASim: Allele-aware spatial RNA-seq simulator enables systematic benchmarking of copy number inference
Spatial transcriptomics (ST) is revolutionizing the study of tumor evolution by enabling spatially resolved copy-number alteration (CNA) analysis. However, evaluating the accuracy and robustness of current single-cell (SC) and ST-specific CNA inference tools remains challenging due to the absence of ground-truth datasets. Here, we present stCNASim, an allele-aware spatial RNA-seq simulator that generates raw reads within realistic spatial contexts. We synthesized 46 benchmarking datasets across varying technical settings and spatial architectures to evaluate five widely used computational methods. Our analysis reveals that while SC-based methods adapt well to ST data, ST-specific methods successfully benefit from considering spatial autocorrelation but struggle under high spatial intermixing. The allele-aware methods CalicoST, Numbat, and XClone achieved top-tier performance with unique advantages in extreme scenarios, yet showed distinct sensitivities to low purity, mirrored alleles, and low coverage, respectively. By providing a scalable simulator and a rigorous benchmark, this work establishes a much-needed framework to guide and accelerate future tool development in spatial CNA analysis.
- 30 July 2026
- Published by Springer Nature in Nature Computational Science
- Vol. 6 (9) , 971-987
SpatialFormer: universal spatial representation learning from subcellular molecular to multicellular landscapes
- 30 July 2026
- Published by Springer Nature in Nature Computational Science
- Vol. 6 (9) , 971-987
SpatialFormer: universal spatial representation learning from subcellular molecular to multicellular landscapes
- Open Access
- 28 July 2026
- Published by Springer Nature in Molecular Systems Biology
- Vol. 22 (11) , 1672-1687
xDecoder unlocks the potential of genomic foundation models for few-shot personal gene expression prediction
Large-scale genomic language models (gLMs) hold promise for modeling gene regulation, yet their ability to capture personal gene expression variations remains unresolved. We developed xDecoder, a unified decoding framework that utilizes gLMs and sequence-to-function (S2F) embeddings to learn how personal genetic variation shapes gene expression from paired genome-transcriptome data. Compared to the pretrained genomic models, xDecoder with personalized DNA–RNA training makes cross-individual prediction tractable for seen genes in a few-shot setting. However, zero-shot prediction at unseen loci remains unreliable and gene-dependent, revealing a cross-locus transfer bottleneck of current sequence models. Experiments incorporating individual-level chromatin accessibility suggested that regulatory-state information important for unseen-locus prediction is not fully captured by current DNA-only models. Overall, these results highlight the potential utility of the few-shot setting, the limitations of DNA-only models, and point toward multi-omic, variant-aware frameworks as a promising direction for building personalized regulatory models.
- 28 July 2026
- Published by Springer Nature in Molecular Systems Biology
- Vol. 22 (11) , 1672-1687
Open AccessxDecoder unlocks the potential of genomic foundation models for few-shot personal gene expression prediction
Large-scale genomic language models (gLMs) hold promise for modeling gene regulation, yet their ability to capture personal gene expression variations remains unresolved. We developed xDecoder, a unified decoding framework that utilizes gLMs and sequence-to-function (S2F) embeddings to learn how personal genetic variation shapes gene expression from paired genome-transcriptome data. Compared to the pretrained genomic models, xDecoder with personalized DNA–RNA training makes cross-individual prediction tractable for seen genes in a few-shot setting. However, zero-shot prediction at unseen loci remains unreliable and gene-dependent, revealing a cross-locus transfer bottleneck of current sequence models. Experiments incorporating individual-level chromatin accessibility suggested that regulatory-state information important for unseen-locus prediction is not fully captured by current DNA-only models. Overall, these results highlight the potential utility of the few-shot setting, the limitations of DNA-only models, and point toward multi-omic, variant-aware frameworks as a promising direction for building personalized regulatory models.
- Open Access
SHINE: Decoding transcriptional-metabolic microenvironments through higher-order spatial integration
Spatial omics technologies are expanding to co-profile transcriptomics and metabolomics on the same tissue slide, providing complementary views of gene expression and biochemical activity to reveal molecular programs within native tissue microenvironments. However, integrating the transcriptome and metabolome remains technically challenging due to spatial misalignment, resolution disparity, and higher-order cross-modality interactions. Here, we present SHINE , a hypergraph-based computational framework for the joint analysis of spatial gene expression and metabolic networks derived from the co-profiling slide, focusing on representation learning and cross-modality interaction. Across multiple datasets, SHINE consistently outperformed existing methods for domain segmentation and biomarker co-localization, and provided interpretable insights into metabolic–transcriptional microenvironments. Specifically, on Parkinson’s disease mouse models, SHINE accurately delineates dopaminergic neuron-depleted regions and reconstructs coherent dopamine-associated axes. In human lung and breast cancers, SHINE resolves tumor-associated spatial regions and identifies spatially organized gene–metabolite programs associated with the tumor microenvironment. SHINE enables scalable spatial multi-omics integration across diverse biological systems.
Open AccessSHINE: Decoding transcriptional-metabolic microenvironments through higher-order spatial integration
Spatial omics technologies are expanding to co-profile transcriptomics and metabolomics on the same tissue slide, providing complementary views of gene expression and biochemical activity to reveal molecular programs within native tissue microenvironments. However, integrating the transcriptome and metabolome remains technically challenging due to spatial misalignment, resolution disparity, and higher-order cross-modality interactions. Here, we present SHINE , a hypergraph-based computational framework for the joint analysis of spatial gene expression and metabolic networks derived from the co-profiling slide, focusing on representation learning and cross-modality interaction. Across multiple datasets, SHINE consistently outperformed existing methods for domain segmentation and biomarker co-localization, and provided interpretable insights into metabolic–transcriptional microenvironments. Specifically, on Parkinson’s disease mouse models, SHINE accurately delineates dopaminergic neuron-depleted regions and reconstructs coherent dopamine-associated axes. In human lung and breast cancers, SHINE resolves tumor-associated spatial regions and identifies spatially organized gene–metabolite programs associated with the tumor microenvironment. SHINE enables scalable spatial multi-omics integration across diverse biological systems.
- Open Access
BRIDGE: A Multi-organ Histo-ST Foundation Model Enables Virtual Spatial Transcriptomics for Enhanced Few-shot Cancer Diagnosis
SUMMARY: Recent studies have explored generating virtual spatial transcriptomics (ST) profiles from histological images, offering a promising alternative to laboratory-measured molecular profiling. However, existing approaches predominantly rely on single-organ models and require sub-stantial organ-specific training data, restricting their accuracy under challenging few-shot conditions in clincical practice, where less than 10 slides are available for specific organs or techniques. Here, we present BRIDGE, a multi-organ foundation model pre-trained on over 600,000 paired histology-ST profiles across 13 human organs and three sequencing techniques. By robustly aligning morphological features and genomic information within a shared multi-organ latent space, BRIDGE can leverage common biological knowledge across distinct tissues to enable accurate and generalizable pan-cancer molecular profiling. Without additional organ-specific fine-tuning, BRIDGE accurately predicts the spatial expression of 80 biomarker genes, achieving an average Pearson correlation coefficient (PCC) of 0.474—a 30% improvement over existing state-of-the-art models under three clinically challenging few-shot scenarios. With generated virtual ST, BRIDGE outperforms current state-of-the-art pathology foundation models in predicting cancer survival, achieving an average concordance index (C-index) of 0.724 across six TCGA cohorts. Notably, BRIDGE maintains exceptional performance even in zero-shot scenarios involving three cancer types not seen during its training, achieving an average C-index of 0.717, thereby demonstrating its strong generalization capability that transcends organ- and subtype-specific boundaries. Moreover, BRIDGE-generated virtual spatial transcriptomes match the prognostic accuracy of bulk RNA-seq, highlighting their potential as a spatially informative alternative to laboratory sequencing. In general, BRIDGE represents a data-efficient tool in virtual ST that facilitates biomedical discoveries in clinical few-shot contexts and advances diagnosis of understudied cancers without sufficient samples.
Open AccessBRIDGE: A Multi-organ Histo-ST Foundation Model Enables Virtual Spatial Transcriptomics for Enhanced Few-shot Cancer Diagnosis
SUMMARY: Recent studies have explored generating virtual spatial transcriptomics (ST) profiles from histological images, offering a promising alternative to laboratory-measured molecular profiling. However, existing approaches predominantly rely on single-organ models and require sub-stantial organ-specific training data, restricting their accuracy under challenging few-shot conditions in clincical practice, where less than 10 slides are available for specific organs or techniques. Here, we present BRIDGE, a multi-organ foundation model pre-trained on over 600,000 paired histology-ST profiles across 13 human organs and three sequencing techniques. By robustly aligning morphological features and genomic information within a shared multi-organ latent space, BRIDGE can leverage common biological knowledge across distinct tissues to enable accurate and generalizable pan-cancer molecular profiling. Without additional organ-specific fine-tuning, BRIDGE accurately predicts the spatial expression of 80 biomarker genes, achieving an average Pearson correlation coefficient (PCC) of 0.474—a 30% improvement over existing state-of-the-art models under three clinically challenging few-shot scenarios. With generated virtual ST, BRIDGE outperforms current state-of-the-art pathology foundation models in predicting cancer survival, achieving an average concordance index (C-index) of 0.724 across six TCGA cohorts. Notably, BRIDGE maintains exceptional performance even in zero-shot scenarios involving three cancer types not seen during its training, achieving an average C-index of 0.717, thereby demonstrating its strong generalization capability that transcends organ- and subtype-specific boundaries. Moreover, BRIDGE-generated virtual spatial transcriptomes match the prognostic accuracy of bulk RNA-seq, highlighting their potential as a spatially informative alternative to laboratory sequencing. In general, BRIDGE represents a data-efficient tool in virtual ST that facilitates biomedical discoveries in clinical few-shot contexts and advances diagnosis of understudied cancers without sufficient samples.
- Open Access
- 5 May 2026
- Published by Springer Nature in Genome Biology
- Vol. 27
CRAK-Velo: chromatin accessibility kinetics integration improves RNA velocity estimation
RNA velocity has recently emerged as a key tool in the analysis of single-cell transcriptomic data, yet connecting RNA velocity analyses to underlying regu
- 5 May 2026
- Published by Springer Nature in Genome Biology
- Vol. 27
Open AccessCRAK-Velo: chromatin accessibility kinetics integration improves RNA velocity estimation
RNA velocity has recently emerged as a key tool in the analysis of single-cell transcriptomic data, yet connecting RNA velocity analyses to underlying regu
- Open Access
SCIMETAR-seq tracks immunophenotype, demethylation, mutations, and transcriptomes in single cells undergoing HMA therapy
Myelodysplastic neoplasms (MDS) and related myeloid neoplasms such as chronic myelomonocytic leukaemia (CMML) are clonal haematopoietic stem cell disorders characterised by ineffective and dysplastic haematopoiesis. They are associated with peripheral cytopaenias, variable increases in immature blasts, and a risk of progression to acute myeloid leukaemia. Hypomethylating agents (HMA) can improve blood counts and reduce blasts, but responses are usually limited. Epigenetic rewiring of haematopoietic stem and progenitor cells (HSPC) by HMA enhances hematopoietic output but is influenced by clonal mosaicism, which requires tracking of response at the single cell level to achieve full understanding. We developed SCIMETAR-seq for s ingle- c ell interrogation of DNA me thylation, target a mplicons, and m R NA in FACS-indexed HSPC, then deployed SCIMETAR-seq on CD34 + HSPC from longitudinal HMA-treated patient BM in vitro and in vivo . HMA-induced LINE-1 (L1) demethylation was positively correlated with cell cycling; being lowest in quiescent HSC and highest in erythrocyte progenitors. Erythrocyte progenitor frequencies were particularly increased by HMA exposure. SRSF2 p.P95 genotype did not influence HMA-induced L1 demethylation but was enriched into cells with a CMP immunophenotype, which were transcriptionally biased away from MEP towards granulocytic progenitors. Despite a lack of L1 demethylation in quiescent HSC/MPP after 7 days of HMA treatment in vivo , their transcriptomes were enriched for TNFα-, TGFβ- and WNT-signaling, suggesting that extrinsic factors secreted by other BM cells in response to HMA mediates reprogramming of quiescent HSC during HMA therapy in vivo .
Open AccessSCIMETAR-seq tracks immunophenotype, demethylation, mutations, and transcriptomes in single cells undergoing HMA therapy
Myelodysplastic neoplasms (MDS) and related myeloid neoplasms such as chronic myelomonocytic leukaemia (CMML) are clonal haematopoietic stem cell disorders characterised by ineffective and dysplastic haematopoiesis. They are associated with peripheral cytopaenias, variable increases in immature blasts, and a risk of progression to acute myeloid leukaemia. Hypomethylating agents (HMA) can improve blood counts and reduce blasts, but responses are usually limited. Epigenetic rewiring of haematopoietic stem and progenitor cells (HSPC) by HMA enhances hematopoietic output but is influenced by clonal mosaicism, which requires tracking of response at the single cell level to achieve full understanding. We developed SCIMETAR-seq for s ingle- c ell interrogation of DNA me thylation, target a mplicons, and m R NA in FACS-indexed HSPC, then deployed SCIMETAR-seq on CD34 + HSPC from longitudinal HMA-treated patient BM in vitro and in vivo . HMA-induced LINE-1 (L1) demethylation was positively correlated with cell cycling; being lowest in quiescent HSC and highest in erythrocyte progenitors. Erythrocyte progenitor frequencies were particularly increased by HMA exposure. SRSF2 p.P95 genotype did not influence HMA-induced L1 demethylation but was enriched into cells with a CMP immunophenotype, which were transcriptionally biased away from MEP towards granulocytic progenitors. Despite a lack of L1 demethylation in quiescent HSC/MPP after 7 days of HMA treatment in vivo , their transcriptomes were enriched for TNFα-, TGFβ- and WNT-signaling, suggesting that extrinsic factors secreted by other BM cells in response to HMA mediates reprogramming of quiescent HSC during HMA therapy in vivo .
- Top Cited PapersOpen Access
- 16 March 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
FineST: contrastive learning integrates histology and spatial transcriptomics for nuclei-resolved ligand-receptor analysis
Spatial transcriptomics (ST) has emerged as a powerful tool for analyzing cell-cell communication (CCC) across various biological processes, ranging from embryonic development to cancer progression. However, its limited resolution and high data sparsity hinder the detailed characterization of CCC patterns within complex tissues. Here, we introduce FineST , a deep contrastive learning model that leverages a histology foundation model to fuse ST and histology images, enabling Fine -grained S patial T ranscriptomics analysis. This approach facilitates precise nuclei segmentation, high-resolution RNA expression imputation, and the identification of intricate ligand-receptor interactions. Using both colorectal cancer VisiumHD and breast cancer Xenium datasets, we demonstrate that FineST significantly outperforms existing methods in high-resolution RNA imputation, cell type prediction, and CCC pattern discovery. With focused application to the Visium platform, FineST reveals novel biological insights into tumor-immune interactions across multiple cancer types, including invasive fronts in breast cancer, tertiary lymphoid structures in nasopharyngeal carcinoma, and PD-1 therapy resistance barriers in hepatocellular carcinoma. These findings highlight a new paradigm in ST analysis through the integration of readily available histology images.
- 16 March 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Top Cited PapersOpen AccessFineST: contrastive learning integrates histology and spatial transcriptomics for nuclei-resolved ligand-receptor analysis
Spatial transcriptomics (ST) has emerged as a powerful tool for analyzing cell-cell communication (CCC) across various biological processes, ranging from embryonic development to cancer progression. However, its limited resolution and high data sparsity hinder the detailed characterization of CCC patterns within complex tissues. Here, we introduce FineST , a deep contrastive learning model that leverages a histology foundation model to fuse ST and histology images, enabling Fine -grained S patial T ranscriptomics analysis. This approach facilitates precise nuclei segmentation, high-resolution RNA expression imputation, and the identification of intricate ligand-receptor interactions. Using both colorectal cancer VisiumHD and breast cancer Xenium datasets, we demonstrate that FineST significantly outperforms existing methods in high-resolution RNA imputation, cell type prediction, and CCC pattern discovery. With focused application to the Visium platform, FineST reveals novel biological insights into tumor-immune interactions across multiple cancer types, including invasive fronts in breast cancer, tertiary lymphoid structures in nasopharyngeal carcinoma, and PD-1 therapy resistance barriers in hepatocellular carcinoma. These findings highlight a new paradigm in ST analysis through the integration of readily available histology images.