Can Yang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Can Yang has a total of 189 co-authors. Here are the top co-authors by number of shared publications:
Jiashun Xiao
HKUST
17 shared publications- Jin Liu
The Chinese University of Hong Kong, Shenzhen
15 shared publications Mingxuan Cai
Department of Biostatistics, City University of Hong Kong, Hong Kong, China
13 shared publicationsXianghong Hu
Shenzhen University
11 shared publicationsYang Wang
Department of Mathematics, The Hong Kong University of Science and Technology, Hong Kong SAR, China
11 shared publications- Angela Ruohao Wu
The Hong Kong University of Science and Technology
10 shared publications
Latest Publications
- Open Access
MRCAT: A multi-agent framework for reliable Mendelian randomization analysis and causal interpretation
Mendelian randomization (MR) uses genetic variants to estimate causal effects of exposures on health outcomes, but rigorous MR analyses require coordinated procedures and interpretation under explicit methodological assumptions. Although artificial intelligence (AI) agents can automate analyses, computational completion does not guarantee scientific reliability. Here, we present MRCAT, a multi-agent framework that translates MR-specific knowledge into an AI-assisted workflow that is plannable, executable, verifiable, recoverable and traceable. MRCAT combines dependency-aware planning, Skill-guided execution and expert-informed review with traceable reporting guided by STROBE-MR, a guideline for clear and transparent reporting of MR studies, and iterative Writer-Judge checking. We evaluated MRCAT on 30 natural-language queries spanning diverse data sources, analytical requirements and boundary conditions. MRCAT achieved scientific completion, defined as meeting all nine prespecified MR-specific procedural requirements, for all 26 executable queries and appropriately handled all four boundary cases, compared with 17, 11 and 4 scientifically completed queries for Claude Code, Biomni and BioMedAgent, respectively. Across six independently curated tasks for which the expert-curated analyses did not support causal effects, MRCAT showed closer numerical concordance with the expert-curated analyses and produced interpretations consistent with expert-curated conclusions in all six cases, compared with two for Claude Code and four for Biomni. Additional case studies demonstrated evidence-grounded reporting and selective reuse of analytical outputs in multi-exposure and bidirectional MR investigations. MRCAT provides a framework for reducing the operational burden of MR research while supporting reproducible analysis, assumption-consistent interpretation and traceable scientific reporting.
Open AccessMRCAT: A multi-agent framework for reliable Mendelian randomization analysis and causal interpretation
Mendelian randomization (MR) uses genetic variants to estimate causal effects of exposures on health outcomes, but rigorous MR analyses require coordinated procedures and interpretation under explicit methodological assumptions. Although artificial intelligence (AI) agents can automate analyses, computational completion does not guarantee scientific reliability. Here, we present MRCAT, a multi-agent framework that translates MR-specific knowledge into an AI-assisted workflow that is plannable, executable, verifiable, recoverable and traceable. MRCAT combines dependency-aware planning, Skill-guided execution and expert-informed review with traceable reporting guided by STROBE-MR, a guideline for clear and transparent reporting of MR studies, and iterative Writer-Judge checking. We evaluated MRCAT on 30 natural-language queries spanning diverse data sources, analytical requirements and boundary conditions. MRCAT achieved scientific completion, defined as meeting all nine prespecified MR-specific procedural requirements, for all 26 executable queries and appropriately handled all four boundary cases, compared with 17, 11 and 4 scientifically completed queries for Claude Code, Biomni and BioMedAgent, respectively. Across six independently curated tasks for which the expert-curated analyses did not support causal effects, MRCAT showed closer numerical concordance with the expert-curated analyses and produced interpretations consistent with expert-curated conclusions in all six cases, compared with two for Claude Code and four for Biomni. Additional case studies demonstrated evidence-grounded reporting and selective reuse of analytical outputs in multi-exposure and bidirectional MR investigations. MRCAT provides a framework for reducing the operational burden of MR research while supporting reproducible analysis, assumption-consistent interpretation and traceable scientific reporting.
- Open Access
Hierarchical distribution matching enables comprehensive characterization of common and condition-specific cell niches in spatial omics data
Deciphering cell niches in complex tissues is essential for understanding tissue structure and disease. Recent advances in spatial omics have enabled subcellular resolution and accurate cell identity mapping. However, robust delineation of cell niches and disease-associated spatial patterns remains difficult. We introduce Harmonics, a novel computational framework that systematically identifies both common and condition-specific cell niches from spatial omics data through hierarchical distribution matching. Harmonics also includes a suite of downstream modules that facilitate comprehensive niche characterization. We demonstrate its scalability, accuracy and generalizability across datasets spanning diverse species, tissues, diseases, spatial modalities, and technological platforms. For condition-agnostic datasets, Harmonics outperforms baseline methods in both accuracy and robustness, and further demonstrates the capability to resolve niche structures at finer granularity. Across diverse diseases including pulmonary fibrosis, triple-negative breast cancer, and colorectal cancer, Harmonics enables precise identification of condition-specific niches and reveals disease-associated dynamics, subtype-specific spatial patterns, and structured immune architectures. We envision Harmonics as a practical and versatile tool for spatial niche analysis that can be applied across a wide range of biological contexts and seamlessly integrated with existing spatial omics workflows.
Open AccessHierarchical distribution matching enables comprehensive characterization of common and condition-specific cell niches in spatial omics data
Deciphering cell niches in complex tissues is essential for understanding tissue structure and disease. Recent advances in spatial omics have enabled subcellular resolution and accurate cell identity mapping. However, robust delineation of cell niches and disease-associated spatial patterns remains difficult. We introduce Harmonics, a novel computational framework that systematically identifies both common and condition-specific cell niches from spatial omics data through hierarchical distribution matching. Harmonics also includes a suite of downstream modules that facilitate comprehensive niche characterization. We demonstrate its scalability, accuracy and generalizability across datasets spanning diverse species, tissues, diseases, spatial modalities, and technological platforms. For condition-agnostic datasets, Harmonics outperforms baseline methods in both accuracy and robustness, and further demonstrates the capability to resolve niche structures at finer granularity. Across diverse diseases including pulmonary fibrosis, triple-negative breast cancer, and colorectal cancer, Harmonics enables precise identification of condition-specific niches and reveals disease-associated dynamics, subtype-specific spatial patterns, and structured immune architectures. We envision Harmonics as a practical and versatile tool for spatial niche analysis that can be applied across a wide range of biological contexts and seamlessly integrated with existing spatial omics workflows.
- Open Access
- 14 October 2026
- Published by Springer Nature in Nature Communications
- Vol. 17  (1)
Mapping cell-type-specific spatial gene programs uncovers tissue architecture and microenvironment organization
The complex organization of tissues is shaped by the spatial arrangement of diverse cell types executing specialized gene programs. Cell-type-specific spatial gene programs (SGPs) are transcriptional modules that capture spatial expression co-variation within cell types. However, identifying these SGPs remains challenging in spatial transcriptomics. Here we present MacSGP, a scalable framework that integrates deep graph neural networks with probabilistic modeling to identify cell-type-specific SGPs through low-rank structures while accounting for cell type mixtures. Through comprehensive simulations and real data analyses spanning diverse tissues and technologies, we demonstrate that MacSGP effectively distinguishes cell-type-specific SGPs from cell-type marker gene expression patterns while uncovering tissue architecture and cellular microenvironments. In a mouse brain dataset, we identify replicable SGPs reflecting subregional neuronal specializations, with these genes significantly enriched for heritabilities associated with brain-related phenotypes. Across colorectal cancer datasets from various platforms, MacSGP identifies consistent SGPs that define distinct cellular niches within the tumor microenvironment.
- 14 October 2026
- Published by Springer Nature in Nature Communications
- Vol. 17  (1)
Open AccessMapping cell-type-specific spatial gene programs uncovers tissue architecture and microenvironment organization
The complex organization of tissues is shaped by the spatial arrangement of diverse cell types executing specialized gene programs. Cell-type-specific spatial gene programs (SGPs) are transcriptional modules that capture spatial expression co-variation within cell types. However, identifying these SGPs remains challenging in spatial transcriptomics. Here we present MacSGP, a scalable framework that integrates deep graph neural networks with probabilistic modeling to identify cell-type-specific SGPs through low-rank structures while accounting for cell type mixtures. Through comprehensive simulations and real data analyses spanning diverse tissues and technologies, we demonstrate that MacSGP effectively distinguishes cell-type-specific SGPs from cell-type marker gene expression patterns while uncovering tissue architecture and cellular microenvironments. In a mouse brain dataset, we identify replicable SGPs reflecting subregional neuronal specializations, with these genes significantly enriched for heritabilities associated with brain-related phenotypes. Across colorectal cancer datasets from various platforms, MacSGP identifies consistent SGPs that define distinct cellular niches within the tumor microenvironment.
- 5 October 2026
- Published by Springer Nature in Nature Computational Science
- p. 1-8
CellART: a unified framework for extracting single-cell information from high-resolution spatial transcriptomics
Recent advances in spatial transcriptomics (ST) have achieved subcellular spatial resolution, yet existing platforms either capture sparse transcript counts per spot or measure only a limited number of genes, complicating the extraction of comprehensive single-cell information. Here we introduce CellART, a unified framework for extracting single-cell information across diverse high-resolution ST platforms. By leveraging multimodal data, including staining images, ST data and single-cell RNA sequencing references, CellART simultaneously performs cell segmentation and cell-type annotation through integration of deep learning and probabilistic modeling. CellART is efficient, generalizable and robust, and its outputs are compatible with widely used community tools, facilitating a variety of downstream analyses.
- 5 October 2026
- Published by Springer Nature in Nature Computational Science
- p. 1-8
CellART: a unified framework for extracting single-cell information from high-resolution spatial transcriptomics
Recent advances in spatial transcriptomics (ST) have achieved subcellular spatial resolution, yet existing platforms either capture sparse transcript counts per spot or measure only a limited number of genes, complicating the extraction of comprehensive single-cell information. Here we introduce CellART, a unified framework for extracting single-cell information across diverse high-resolution ST platforms. By leveraging multimodal data, including staining images, ST data and single-cell RNA sequencing references, CellART simultaneously performs cell segmentation and cell-type annotation through integration of deep learning and probabilistic modeling. CellART is efficient, generalizable and robust, and its outputs are compatible with widely used community tools, facilitating a variety of downstream analyses.
- Open Access
Accurate and scalable decontamination of imaging-based spatial transcriptomics via optimal transport
Imaging-based spatial transcriptomics enables molecule-resolved profiling of gene expression and tissue organization in situ. However, segmentation errors, transcript spillover and three-dimensional cell overlap can introduce misassigned transcripts into cell-level expression profiles, compromising biological interpretation and obscuring genuine signals. Existing methods either remove suspect expression at the cost of signal loss or lack a biologically grounded criterion for transcript assignment. Here we present CellDot, an optimal-transport framework that determines the fate of each transcript by retaining it in its host cell, reassigning it to a plausible neighboring cell or removing it as background. By integrating reference-guided expression compatibility with spatial information and data-adaptive constraints, CellDot enables accurate and traceable molecule-level correction while preserving biologically meaningful variation. In evaluations across multiple human tumor datasets, CellDot exhibited superior performance compared to existing decontamination methods, successfully restoring spatial expression patterns that matched independent cross-platform measurements. Moreover, it significantly enhanced the recovery of cellular states, intercellular communication, and spatial niche programs. Our experiments using real data demonstrated CellDot's scalability and established it as the only method applicable to a whole-transcriptome Atera dataset, underscoring its distinct advantages in the field of spatial transcriptomics.
Open AccessAccurate and scalable decontamination of imaging-based spatial transcriptomics via optimal transport
Imaging-based spatial transcriptomics enables molecule-resolved profiling of gene expression and tissue organization in situ. However, segmentation errors, transcript spillover and three-dimensional cell overlap can introduce misassigned transcripts into cell-level expression profiles, compromising biological interpretation and obscuring genuine signals. Existing methods either remove suspect expression at the cost of signal loss or lack a biologically grounded criterion for transcript assignment. Here we present CellDot, an optimal-transport framework that determines the fate of each transcript by retaining it in its host cell, reassigning it to a plausible neighboring cell or removing it as background. By integrating reference-guided expression compatibility with spatial information and data-adaptive constraints, CellDot enables accurate and traceable molecule-level correction while preserving biologically meaningful variation. In evaluations across multiple human tumor datasets, CellDot exhibited superior performance compared to existing decontamination methods, successfully restoring spatial expression patterns that matched independent cross-platform measurements. Moreover, it significantly enhanced the recovery of cellular states, intercellular communication, and spatial niche programs. Our experiments using real data demonstrated CellDot's scalability and established it as the only method applicable to a whole-transcriptome Atera dataset, underscoring its distinct advantages in the field of spatial transcriptomics.
- Open Access
CellART: a unified framework for extracting single-cell information from high-resolution spatial transcriptomics
Understanding how different cell types assemble into tissues and organs, as well as how they interact to transmit and receive biological signals, is essential for advancing biomedical and biological research. Recent advancements in spatial transcriptomics (ST) technologies have opened new avenues for investigating biological systems by achieving subcellular spatial resolution. Since cells are the fundamental units of life, extracting single-cell information from high-resolution ST data is crucial. However, existing ST platforms often capture sparse transcript counts per spot or measure only a limited number of genes, complicating the extraction of comprehensive single-cell information. In this study, we introduce CellART, a unified framework designed to extract single-cell information across diverse high-resolution ST platforms, including VisiumHD, Xenium, MERFISH, and Stereo-seq. By leveraging multimodal data, such as staining images, spatial transcriptomics data, and single-cell RNA sequencing references, CellART simultaneously performs cell segmentation and cell type annotation through a seamless integration of deep learning and probabilistic modeling. We demonstrate the efficiency, generalizability, and robustness of CellART across various high-resolution spatial transcriptomics platforms, capable of processing datasets containing millions of spots. Comprehensive experiments validate the biological relevance and accuracy of the recovered cellular information within spatial configurations. Notably, we highlight the utility of CellART in breast and colorectal cancer datasets, showcasing its ability to fully leverage high-resolution ST data. By enhancing cellular resolution, CellART facilitates the identification of transient cancer cell states and immune cell subtypes. Furthermore, CellART enables investigations into cancer-immune cell communication, uncovering both established interactions and novel ligand-receptor pairs. The outputs of CellART are compatible with widely used community tools, facilitating a variety of downstream analyses.
Open AccessCellART: a unified framework for extracting single-cell information from high-resolution spatial transcriptomics
Understanding how different cell types assemble into tissues and organs, as well as how they interact to transmit and receive biological signals, is essential for advancing biomedical and biological research. Recent advancements in spatial transcriptomics (ST) technologies have opened new avenues for investigating biological systems by achieving subcellular spatial resolution. Since cells are the fundamental units of life, extracting single-cell information from high-resolution ST data is crucial. However, existing ST platforms often capture sparse transcript counts per spot or measure only a limited number of genes, complicating the extraction of comprehensive single-cell information. In this study, we introduce CellART, a unified framework designed to extract single-cell information across diverse high-resolution ST platforms, including VisiumHD, Xenium, MERFISH, and Stereo-seq. By leveraging multimodal data, such as staining images, spatial transcriptomics data, and single-cell RNA sequencing references, CellART simultaneously performs cell segmentation and cell type annotation through a seamless integration of deep learning and probabilistic modeling. We demonstrate the efficiency, generalizability, and robustness of CellART across various high-resolution spatial transcriptomics platforms, capable of processing datasets containing millions of spots. Comprehensive experiments validate the biological relevance and accuracy of the recovered cellular information within spatial configurations. Notably, we highlight the utility of CellART in breast and colorectal cancer datasets, showcasing its ability to fully leverage high-resolution ST data. By enhancing cellular resolution, CellART facilitates the identification of transient cancer cell states and immune cell subtypes. Furthermore, CellART enables investigations into cancer-immune cell communication, uncovering both established interactions and novel ligand-receptor pairs. The outputs of CellART are compatible with widely used community tools, facilitating a variety of downstream analyses.
- Open Access
- 5 September 2026
- Published by Springer Nature in Nature Communications
- Vol. 17  (1)
Benchmarking copy number alteration inference methods for spatial transcriptomics
Copy number alterations (CNAs), gains or losses of genomic regions, contribute to malignant progression and tumor heterogeneity. Advances in spatial transcriptomics have expanded opportunities to study clonal structure in situ, but direct spatial genomic profiling remains difficult in practice, motivating the increasing use of computational methods to infer CNAs from spatial transcriptomics data. However, their performance across diverse spatial transcriptomics settings remains unclear. Here, we present a benchmark of nine CNA inference methods across 69 spatial transcriptomics tissue sections from six cancer types and four spatial transcriptomics platforms. By evaluating these methods across four key tasks, we show that no single method consistently outperforms all others, with performance depending on the analytical goal and data characteristics. We therefore provide task-specific and data-aware guidance to help users select appropriate methods in practical settings. More broadly, this benchmark provides a basis for the future development and optimization of CNA inference methods.
- 5 September 2026
- Published by Springer Nature in Nature Communications
- Vol. 17  (1)
Open AccessBenchmarking copy number alteration inference methods for spatial transcriptomics
Copy number alterations (CNAs), gains or losses of genomic regions, contribute to malignant progression and tumor heterogeneity. Advances in spatial transcriptomics have expanded opportunities to study clonal structure in situ, but direct spatial genomic profiling remains difficult in practice, motivating the increasing use of computational methods to infer CNAs from spatial transcriptomics data. However, their performance across diverse spatial transcriptomics settings remains unclear. Here, we present a benchmark of nine CNA inference methods across 69 spatial transcriptomics tissue sections from six cancer types and four spatial transcriptomics platforms. By evaluating these methods across four key tasks, we show that no single method consistently outperforms all others, with performance depending on the analytical goal and data characteristics. We therefore provide task-specific and data-aware guidance to help users select appropriate methods in practical settings. More broadly, this benchmark provides a basis for the future development and optimization of CNA inference methods.
- Open Access
- 4 September 2026
- Published by Springer Nature in Research Square
Generative atlasing in universal gene expression space defines cell types and microenvironment spectra during disease progression
Cells execute core physiological functions through intricate molecular networks. Elucidating these network principles is critical for understanding disease progression. However, most current transformer-based single-cell foundation models rely on encoder-only architectures that output abstract latent embeddings lacking explicit biological interpretability. An effective generative framework must quantitatively map these latent coordinates back into the biological real space of gene expression. To achieve this, we present UniGeneX, a generative single-cell foundation model that reconstructs large-scale, multi-dataset transcriptomics data into a quantitative space of Universal Gene Expression (UGE). By constructing the UGE space, UniGeneX effectively disentangles the molecular spectra of cell states across health and disease. This generative framework seamlessly integrates single-cell cohorts and spatial transcriptomics (ST) data, facilitating cell-type identification and niche architecture mapping. Applying UniGeneX to pulmonary fibrosis (PF) and glioma cohorts, we demonstrate its utility in constructing comprehensive tissue atlases, tracking disease progression along a spatial niche pseudotime, and identifying localized vascular zonation programs associated with tumor cellular state transitions.
- 4 September 2026
- Published by Springer Nature in Research Square
Open AccessGenerative atlasing in universal gene expression space defines cell types and microenvironment spectra during disease progression
Cells execute core physiological functions through intricate molecular networks. Elucidating these network principles is critical for understanding disease progression. However, most current transformer-based single-cell foundation models rely on encoder-only architectures that output abstract latent embeddings lacking explicit biological interpretability. An effective generative framework must quantitatively map these latent coordinates back into the biological real space of gene expression. To achieve this, we present UniGeneX, a generative single-cell foundation model that reconstructs large-scale, multi-dataset transcriptomics data into a quantitative space of Universal Gene Expression (UGE). By constructing the UGE space, UniGeneX effectively disentangles the molecular spectra of cell states across health and disease. This generative framework seamlessly integrates single-cell cohorts and spatial transcriptomics (ST) data, facilitating cell-type identification and niche architecture mapping. Applying UniGeneX to pulmonary fibrosis (PF) and glioma cohorts, we demonstrate its utility in constructing comprehensive tissue atlases, tracking disease progression along a spatial niche pseudotime, and identifying localized vascular zonation programs associated with tumor cellular state transitions.
- Open Access
Characterizing dynamic tissue architectures by identifying cell-type-specific spatiotemporal gene programs with stGP
Cellular gene programs unfold over biological time within spatially organized tissues. The same cell type can activate distinct programs in different spatial domains or multicellular niches. Spatiotemporal transcriptomics enables in situ measurement of these processes; however, the interplay between temporal progression and spatial organization complicates the identification of whether a gene program is influenced by temporal changes, spatial structure, or both. To overcome this challenge, we present spatiotemporal Gene Programs (stGP), a statistical framework for identifying interpretable cell-type-specific gene programs across multi-sample spatiotemporal transcriptomic studies. stGP preserves the molecular identity of each program through shared gene loadings, and decomposes individual cell activity into a temporal component that captures gene program responses over biological time, and a spatial component that characterizes variations within tissue sections. We quantify their relative contributions by estimating their variance components. Through comprehensive simulations and analyses of three spatiotemporal transcriptomic datasets across different tissues and technologies, stGP reveals spatiotemporal gene programs that distinguish preserved tissue architecture from dynamic remodeling. Our framework delineates age-associated cellular responses and uncovers localized program deployment within anatomical regions, aging hotspots, and multicellular niches. Our results establish stGP as an effective and robust framework for dissecting dynamic tissue architecture, providing insights into how cell-type-specific gene programs are coordinated across biological time and spatial microenvironments.
Open AccessCharacterizing dynamic tissue architectures by identifying cell-type-specific spatiotemporal gene programs with stGP
Cellular gene programs unfold over biological time within spatially organized tissues. The same cell type can activate distinct programs in different spatial domains or multicellular niches. Spatiotemporal transcriptomics enables in situ measurement of these processes; however, the interplay between temporal progression and spatial organization complicates the identification of whether a gene program is influenced by temporal changes, spatial structure, or both. To overcome this challenge, we present spatiotemporal Gene Programs (stGP), a statistical framework for identifying interpretable cell-type-specific gene programs across multi-sample spatiotemporal transcriptomic studies. stGP preserves the molecular identity of each program through shared gene loadings, and decomposes individual cell activity into a temporal component that captures gene program responses over biological time, and a spatial component that characterizes variations within tissue sections. We quantify their relative contributions by estimating their variance components. Through comprehensive simulations and analyses of three spatiotemporal transcriptomic datasets across different tissues and technologies, stGP reveals spatiotemporal gene programs that distinguish preserved tissue architecture from dynamic remodeling. Our framework delineates age-associated cellular responses and uncovers localized program deployment within anatomical regions, aging hotspots, and multicellular niches. Our results establish stGP as an effective and robust framework for dissecting dynamic tissue architecture, providing insights into how cell-type-specific gene programs are coordinated across biological time and spatial microenvironments.
- Open Access
MRMU: A New Paradigm for Mendelian Randomization by Accounting for Measured Covariates and Unmeasured Confounders
Mendelian randomization (MR) is a powerful approach for causal inference, however, its reliability is frequently compromised by unadjusted covariates and unmeasured confounders, such as unmeasured pleiotropy and sample structure. To address these challenges, we introduce MRMU, a novel paradigm for the MR framework. Unlike traditional single-variable or multivariable MR methods, MRMU selects instrumental variables only from the exposure of interest and estimates one exposure effect at a time, while jointly accounting for measured covariates and unmeasured confounders. This design improves the reliability of MR analyses. In simulations and real data, MRMU achieved better type I error control, higher statistical power, and more accurate effect estimation than existing MR methods. Applying to coronary artery disease (CAD), MRMU identified robust cardiometabolic risk factors, including LDL-C, APOB, systolic blood pressure, body mass index, and smoking initiation, with consistent evidence across multiple CAD datasets. In contrast, traits such as HDL-C, height, and educational attainment, which were found to be significant by existing MR methods, were no longer supported by MRMU. MRMU further supported blood pressure-related traits, rather than lipid traits, as the more relevant pathway linking urate to CAD. Finally, by integrating large-scale plasma proteomics data, MRMU identified candidate CAD drug targets beyond established HMGCR- and PCSK9-related pathways, highlighting its utility for therapeutic target prioritization.
Open AccessMRMU: A New Paradigm for Mendelian Randomization by Accounting for Measured Covariates and Unmeasured Confounders
Mendelian randomization (MR) is a powerful approach for causal inference, however, its reliability is frequently compromised by unadjusted covariates and unmeasured confounders, such as unmeasured pleiotropy and sample structure. To address these challenges, we introduce MRMU, a novel paradigm for the MR framework. Unlike traditional single-variable or multivariable MR methods, MRMU selects instrumental variables only from the exposure of interest and estimates one exposure effect at a time, while jointly accounting for measured covariates and unmeasured confounders. This design improves the reliability of MR analyses. In simulations and real data, MRMU achieved better type I error control, higher statistical power, and more accurate effect estimation than existing MR methods. Applying to coronary artery disease (CAD), MRMU identified robust cardiometabolic risk factors, including LDL-C, APOB, systolic blood pressure, body mass index, and smoking initiation, with consistent evidence across multiple CAD datasets. In contrast, traits such as HDL-C, height, and educational attainment, which were found to be significant by existing MR methods, were no longer supported by MRMU. MRMU further supported blood pressure-related traits, rather than lipid traits, as the more relevant pathway linking urate to CAD. Finally, by integrating large-scale plasma proteomics data, MRMU identified candidate CAD drug targets beyond established HMGCR- and PCSK9-related pathways, highlighting its utility for therapeutic target prioritization.