Hong-Dong Li
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
Research Interests
Hong-Dong Li has a total of 126 co-authors. Here are the top co-authors by number of shared publications:
- Jianxin Wang
School of Computer Science and Engineering, Central South University, Changsha, Hunan, China
18 shared publications Cui-Xiang Lin
School of Computer Science and Engineering, Hunan Provincial Key Lab on Bioinformatics , Central South University, Changsha, Hunan 410083
8 shared publications- Jin Liu
School of Computer Science and Engineering, Central South University, Changsha, Hunan, China
6 shared publications Chao Deng
School of Computer Science and Engineering, Central South University , Changsha, 410083, China
4 shared publicationsShaokai Wang
School of Computer Science and Engineering, Central South University , Changsha 410083
4 shared publicationsYunpei Xu
School of Computer Science and Engineering, Central South University , Changsha 410083
4 shared publications
Latest Publications
- 25 September 2026
- Published by Springer Nature
- p. 230-239
DEC-Net: A Multi-task Distillation Expert Collaboration Network for Glioma Grading and Molecular Subtyping Using Multimodal MRI
- 25 September 2026
- Published by Springer Nature
- p. 230-239
DEC-Net: A Multi-task Distillation Expert Collaboration Network for Glioma Grading and Molecular Subtyping Using Multimodal MRI
- Open Access
- 31 August 2026
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (9)
TMEDRP: decoding tumor-intrinsic and microenvironmental signatures for clinical drug response prediction
Motivation: Clinical drug response prediction is constrained by the paucity of patient data, forcing a reliance on in vitro cell line models. Current transfer learning-based methods typically focus on learning domain-invariant representations, but often overlook the critical role of tumor microenvironment (TME) during cross-domain translation. Given that TME discrepancies between in vitro and in vivo settings are critical factors of therapeutic resistance, explicitly modeling the TME is essential to bridge the gap between preclinical models and patient-specific responses. Results: We propose TMEDRP, a novel two-stage learning framework that integrates tumor-intrinsic signatures and TME-specific components for enhanced clinical drug response prediction. TMEDRP introduces TME influences via a pathway-informed disentanglement strategy, a neural co-expression module, and an uncertainty-driven adaptation mechanism. Extensive benchmarks demonstrate that TMEDRP outperforms existing state-of-the-art methods. Importantly, it exhibits promising zero-shot generalization across unseen cancer lineages and novel compounds, suggesting its robustness across diverse scenarios. Furthermore, quantitative analysis of the model’s latent embeddings uncovers key tumor-intrinsic pathways and extrinsic TME factors that distinguish drug-sensitive from resistant cohorts. The identified shared and drug-specific molecular markers align with established clinical mechanisms, supporting the model’s biological interpretability and clinical utility. Availability and implementation: The code and results are available at https://github.com/kybinn/TMEDRP.
- 31 August 2026
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (9)
Open AccessTMEDRP: decoding tumor-intrinsic and microenvironmental signatures for clinical drug response prediction
Motivation: Clinical drug response prediction is constrained by the paucity of patient data, forcing a reliance on in vitro cell line models. Current transfer learning-based methods typically focus on learning domain-invariant representations, but often overlook the critical role of tumor microenvironment (TME) during cross-domain translation. Given that TME discrepancies between in vitro and in vivo settings are critical factors of therapeutic resistance, explicitly modeling the TME is essential to bridge the gap between preclinical models and patient-specific responses. Results: We propose TMEDRP, a novel two-stage learning framework that integrates tumor-intrinsic signatures and TME-specific components for enhanced clinical drug response prediction. TMEDRP introduces TME influences via a pathway-informed disentanglement strategy, a neural co-expression module, and an uncertainty-driven adaptation mechanism. Extensive benchmarks demonstrate that TMEDRP outperforms existing state-of-the-art methods. Importantly, it exhibits promising zero-shot generalization across unseen cancer lineages and novel compounds, suggesting its robustness across diverse scenarios. Furthermore, quantitative analysis of the model’s latent embeddings uncovers key tumor-intrinsic pathways and extrinsic TME factors that distinguish drug-sensitive from resistant cohorts. The identified shared and drug-specific molecular markers align with established clinical mechanisms, supporting the model’s biological interpretability and clinical utility. Availability and implementation: The code and results are available at https://github.com/kybinn/TMEDRP.
- Open Access
- 31 August 2026
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (9)
scMustree: a multi-scale functional hierarchy for single-cell transcriptomic analysis
Motivation: Resolving cellular heterogeneity requires methods that capture both discrete cell types and their functional relationships. Existing single-cell clustering approaches often produce flat partitions, limiting their ability to reveal rare cell states and continuous biological transitions. Here, we introduce scMustree, a computational framework that constructs a multi-scale hierarchy of single-cell transcriptomes, enabling integrated analysis of cellular organization and functional relationships. scMustree employs a top-down iterative decomposition to isolate transcriptionally homogeneous populations, followed by a bottom-up merging strategy guided by cluster-specific functional rankings—derived from differential expression and an isolation-forest-based scoring mechanism that quantifies functional distinctness. This unified approach captures lineage structures, functional similarities, and transitional states. Results: Across 11 benchmark datasets, scMustree achieves competitive clustering accuracy while offering substantially enhanced biological interpretability. In diverse biological systems, it uncovers biologically consistent hierarchies and identifies previously uncharacterized cell states, including fibroblast and T-cell subsets in cancer and lipid-associated microglial populations in Alzheimer’s disease. By integrating structural and functional information, scMustree provides a scalable and biologically grounded framework for multi-resolution exploration of single-cell ecosystems, enabling discovery of rare and disease-relevant cell states across diverse biological contexts. Availability and implementation: Freely available at Github (https://github.com/xuyp-csu/scMustree) and Zenodo (https://zenodo.org/records/17480562).
- 31 August 2026
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (9)
Open AccessscMustree: a multi-scale functional hierarchy for single-cell transcriptomic analysis
Motivation: Resolving cellular heterogeneity requires methods that capture both discrete cell types and their functional relationships. Existing single-cell clustering approaches often produce flat partitions, limiting their ability to reveal rare cell states and continuous biological transitions. Here, we introduce scMustree, a computational framework that constructs a multi-scale hierarchy of single-cell transcriptomes, enabling integrated analysis of cellular organization and functional relationships. scMustree employs a top-down iterative decomposition to isolate transcriptionally homogeneous populations, followed by a bottom-up merging strategy guided by cluster-specific functional rankings—derived from differential expression and an isolation-forest-based scoring mechanism that quantifies functional distinctness. This unified approach captures lineage structures, functional similarities, and transitional states. Results: Across 11 benchmark datasets, scMustree achieves competitive clustering accuracy while offering substantially enhanced biological interpretability. In diverse biological systems, it uncovers biologically consistent hierarchies and identifies previously uncharacterized cell states, including fibroblast and T-cell subsets in cancer and lipid-associated microglial populations in Alzheimer’s disease. By integrating structural and functional information, scMustree provides a scalable and biologically grounded framework for multi-resolution exploration of single-cell ecosystems, enabling discovery of rare and disease-relevant cell states across diverse biological contexts. Availability and implementation: Freely available at Github (https://github.com/xuyp-csu/scMustree) and Zenodo (https://zenodo.org/records/17480562).
- Open Access
- 3 August 2026
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (8)
SIMLINK enables accurate variant pathogenicity prediction through modeling the gene-variant-feature association structure
Motivation: Predicting variant pathogenicity is crucial for clinical genetics. Existing approaches face two primary limitations. First, biologically, data for pathogenicity prediction often lacks explicit modeling of the gene-variant-feature association structure. A single gene can harbor multiple variants, and each variant can be characterized by multiple features intrinsically associated with its parent gene. Current methods fail to explicitly model the gene-variant-feature association, thus limiting their performance. Second, methodologically, the variant-pathogenicity association is often assumed to comprise a linear component alongside a nonlinear one. However, current methods typically do not explicitly model the linear component, often failing to disentangle the linear component that might be better addressed with a linear approach. Results: To overcome these limitations, we introduce simultaneous modeling of linear and nonlinear components of knowledge graph (SIMLINK). This novel approach leverages a knowledge graph to model gene-variant-feature associations and a linear model to isolate the linear component. We begin by constructing a variant-centered knowledge graph, comprising over 8 million triplets, which explicitly models the associations between genes, variants, and features. Subsequently, the linear and nonlinear components are learned using a combination of linear and graph neural networks. We train SIMLINK on ClinVar variants. Benchmarking experiments on independent test sets demonstrate its superior prediction on both missense and synonymous variants compared to state-of-the-art methods, including CADD and AlphaMissense. We evaluate the impact of allele frequencies on prediction performance. Applied to variants implicated in Autism Spectrum Disorder, SIMLINK effectively distinguished between high- and low-confidence variants, and critically, the genes harboring top-ranked variants are highly pathogenic. Availability and implementation: The source code is freely available at https://github.com/Chen-LuWang/SIMLINK.
- 3 August 2026
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (8)
Open AccessSIMLINK enables accurate variant pathogenicity prediction through modeling the gene-variant-feature association structure
Motivation: Predicting variant pathogenicity is crucial for clinical genetics. Existing approaches face two primary limitations. First, biologically, data for pathogenicity prediction often lacks explicit modeling of the gene-variant-feature association structure. A single gene can harbor multiple variants, and each variant can be characterized by multiple features intrinsically associated with its parent gene. Current methods fail to explicitly model the gene-variant-feature association, thus limiting their performance. Second, methodologically, the variant-pathogenicity association is often assumed to comprise a linear component alongside a nonlinear one. However, current methods typically do not explicitly model the linear component, often failing to disentangle the linear component that might be better addressed with a linear approach. Results: To overcome these limitations, we introduce simultaneous modeling of linear and nonlinear components of knowledge graph (SIMLINK). This novel approach leverages a knowledge graph to model gene-variant-feature associations and a linear model to isolate the linear component. We begin by constructing a variant-centered knowledge graph, comprising over 8 million triplets, which explicitly models the associations between genes, variants, and features. Subsequently, the linear and nonlinear components are learned using a combination of linear and graph neural networks. We train SIMLINK on ClinVar variants. Benchmarking experiments on independent test sets demonstrate its superior prediction on both missense and synonymous variants compared to state-of-the-art methods, including CADD and AlphaMissense. We evaluate the impact of allele frequencies on prediction performance. Applied to variants implicated in Autism Spectrum Disorder, SIMLINK effectively distinguished between high- and low-confidence variants, and critically, the genes harboring top-ranked variants are highly pathogenic. Availability and implementation: The source code is freely available at https://github.com/Chen-LuWang/SIMLINK.
- 15 July 2026
- Published by Springer Nature
- p. 53-65
Pathogenicity Prediction of Missense Mutations Through Attention-Based Multi-omics Feature Integration
- 15 July 2026
- Published by Springer Nature
- p. 53-65
Pathogenicity Prediction of Missense Mutations Through Attention-Based Multi-omics Feature Integration
- Top Cited PapersOpen Access
- 18 May 2026
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
- Vol. 58 (12) , 1-41
A Survey: Spatiotemporal Consistency in Video Generation
Video generation aims to produce temporally coherent sequences of visual frames, representing a pivotal advancement in Artificial Intelligence Generated Content (AIGC). Compared to static image generation, video generation poses unique challenges: it demands not only high-quality individual frames but also strong temporal coherence to ensure consistency throughout the spatiotemporal sequence. Although research addressing spatiotemporal consistency in video generation has increased in recent years, systematic reviews focusing on this core issue remain relatively scarce. To fill this gap, this article views the video generation task as a sequential sampling process from a high-dimensional spatiotemporal distribution, and further discusses spatiotemporal consistency. We provide a systematic review of the latest advancements in the field. The content spans multiple dimensions including generation models, feature representations, generation frameworks, post-processing techniques, training strategies, benchmarks and evaluation metrics, with a particular focus on the mechanisms and effectiveness of various methods in maintaining spatiotemporal consistency. Finally, this article explores future research directions and potential challenges in this field, aiming to provide valuable insights for advancing video generation technology. The project link is https://github.com/Yin-Z-Y/A-Survey-Spatiotemporal-Consistency-in-Video-Generation .
- 18 May 2026
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
- Vol. 58 (12) , 1-41
Top Cited PapersOpen AccessA Survey: Spatiotemporal Consistency in Video Generation
Video generation aims to produce temporally coherent sequences of visual frames, representing a pivotal advancement in Artificial Intelligence Generated Content (AIGC). Compared to static image generation, video generation poses unique challenges: it demands not only high-quality individual frames but also strong temporal coherence to ensure consistency throughout the spatiotemporal sequence. Although research addressing spatiotemporal consistency in video generation has increased in recent years, systematic reviews focusing on this core issue remain relatively scarce. To fill this gap, this article views the video generation task as a sequential sampling process from a high-dimensional spatiotemporal distribution, and further discusses spatiotemporal consistency. We provide a systematic review of the latest advancements in the field. The content spans multiple dimensions including generation models, feature representations, generation frameworks, post-processing techniques, training strategies, benchmarks and evaluation metrics, with a particular focus on the mechanisms and effectiveness of various methods in maintaining spatiotemporal consistency. Finally, this article explores future research directions and potential challenges in this field, aiming to provide valuable insights for advancing video generation technology. The project link is https://github.com/Yin-Z-Y/A-Survey-Spatiotemporal-Consistency-in-Video-Generation .
- Open Access
- 1 February 2026
- Published by Tsinghua University Press in Tsinghua Science and Technology
- Vol. 31 (1) , 658-673
Traditional Chinese Medicine Prescription Recommendation for Alzheimer’s Disease Based on Network Propagation and Reinforcement Learning
As the global population ages, the prevalence of Alzheimer’s disease (AD) has been steadily increasing. Traditional Chinese medicine, with abundant ingredients and multi-targeting, has shown promising anti-AD effects. However, the complex mechanism of herbal actions makes it challenging to discover effective herbal prescriptions for AD. In this study, we propose an herbal prescription recommendation approach for AD based on network propagation and reinforcement learning. A target-ingredient-herb network is constructed, and network propagation is used to obtain the network score (Nscore) of an herb. The empirical score (Escore) of the herb pair is calculated based on known prescriptions. The herbal prescription score (HPscore) is then calculated based on Nscore and Escore. Finally, reinforcement learning is combined with HPscore to infer effective prescriptions. By validating the predicted prescriptions, we identify the targets of the herbs and AD-related proteins from the database, refining these results through cross-analysis. Core targets are determined using protein-protein interaction network analysis. GO and KEGG analyses explore biological roles of core targets. Molecular docking confirms interactions between actives and targets. Results show optimal herbal formulations act effectively on targets. These may offer strategies to optimize prescriptions, therapies, and drug development.
- 1 February 2026
- Published by Tsinghua University Press in Tsinghua Science and Technology
- Vol. 31 (1) , 658-673
Open AccessTraditional Chinese Medicine Prescription Recommendation for Alzheimer’s Disease Based on Network Propagation and Reinforcement Learning
As the global population ages, the prevalence of Alzheimer’s disease (AD) has been steadily increasing. Traditional Chinese medicine, with abundant ingredients and multi-targeting, has shown promising anti-AD effects. However, the complex mechanism of herbal actions makes it challenging to discover effective herbal prescriptions for AD. In this study, we propose an herbal prescription recommendation approach for AD based on network propagation and reinforcement learning. A target-ingredient-herb network is constructed, and network propagation is used to obtain the network score (Nscore) of an herb. The empirical score (Escore) of the herb pair is calculated based on known prescriptions. The herbal prescription score (HPscore) is then calculated based on Nscore and Escore. Finally, reinforcement learning is combined with HPscore to infer effective prescriptions. By validating the predicted prescriptions, we identify the targets of the herbs and AD-related proteins from the database, refining these results through cross-analysis. Core targets are determined using protein-protein interaction network analysis. GO and KEGG analyses explore biological roles of core targets. Molecular docking confirms interactions between actives and targets. Results show optimal herbal formulations act effectively on targets. These may offer strategies to optimize prescriptions, therapies, and drug development.
- Open Access
- 27 December 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (1)
Mutual learning for joint disease detection and severity prediction reveals multimodal pathogenesis for neurodegenerative disorders
Motivation: Neurodegenerative disorders influence millions of people worldwide, and uncovering the pathogenesis is of urgent need. Many efforts have been made to detect or predict neurodegenerative disorders, while exploring the pathogenesis has been ignored from a systemic perspective. Results: To handle this issue, we propose a novel and powerful method, referred to as Pathogenesis-aware Mutual-Assistance Classification and Regression Optimization (Pa-MACRO). First, Pa-MACRO incorporates a mutual-assistance bidirectional mapping technique with a joint-embedding fine-grained interpretability module. This can extract the intrinsic factors and their interactions of multimodal pathogenesis. Second, our method can simultaneously classify an at-risk individual and predict the severity triggered by neurodegenerative disorders. Furthermore, to address the small sample size issue and the high-dimensional issue, we meticulously incorporate a semi-supervised cooperative learning method to integrate unlabeled data and extend it to a chromosome-wide setting in the spirit of divide-and-conquer. The Alzheimer’s Disease Neuroimaging Initiative (ADNI) database was used to evaluate Pa-MACRO. Without bells and whistles, Pa-MACRO establishes new state-of-the-art results in various settings while maintaining superior interpretability, verifying its power and versatility in revealing the pathogenesis of neurodegenerative disorders. Availability and implementation: The software is publicly available at https://github.com/ZJ-Techie/Pa-MACRO.
- 27 December 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 42 (1)
Open AccessMutual learning for joint disease detection and severity prediction reveals multimodal pathogenesis for neurodegenerative disorders
Motivation: Neurodegenerative disorders influence millions of people worldwide, and uncovering the pathogenesis is of urgent need. Many efforts have been made to detect or predict neurodegenerative disorders, while exploring the pathogenesis has been ignored from a systemic perspective. Results: To handle this issue, we propose a novel and powerful method, referred to as Pathogenesis-aware Mutual-Assistance Classification and Regression Optimization (Pa-MACRO). First, Pa-MACRO incorporates a mutual-assistance bidirectional mapping technique with a joint-embedding fine-grained interpretability module. This can extract the intrinsic factors and their interactions of multimodal pathogenesis. Second, our method can simultaneously classify an at-risk individual and predict the severity triggered by neurodegenerative disorders. Furthermore, to address the small sample size issue and the high-dimensional issue, we meticulously incorporate a semi-supervised cooperative learning method to integrate unlabeled data and extend it to a chromosome-wide setting in the spirit of divide-and-conquer. The Alzheimer’s Disease Neuroimaging Initiative (ADNI) database was used to evaluate Pa-MACRO. Without bells and whistles, Pa-MACRO establishes new state-of-the-art results in various settings while maintaining superior interpretability, verifying its power and versatility in revealing the pathogenesis of neurodegenerative disorders. Availability and implementation: The software is publicly available at https://github.com/ZJ-Techie/Pa-MACRO.
- Open Access
- 29 August 2025
- Published by Wiley in Advanced Science
- Vol. 12 (38) , e06314
diPaRIS: Dynamic and Interpretable Protein‐RNA Interactions Prediction With U‐Shaped Network and Novel Structure Encoding
Protein‐RNA interactions play a critical role in various biological processes and disease development. Proteins interact with RNAs through dynamic binding sites that exhibit specific structural patterns under various cellular conditions. While current computational methods take RNA structures in vivo into account, they fall short in capturing the structure contextual association of nucleotides, limiting predictive accuracy. Here, diPaRIS, a deep learning method, is proposed to predict dynamic protein‐RNA interactions with improved accuracy and enhanced interpretability by integrating RNA structures in vivo. diPaRIS introduces a novel encoding scheme for SHAPE‐seq to encode nucleotide correlations, providing a more comprehensive representation of RNA structures. Leveraging a U‐shaped network architecture, diPaRIS not only improves prediction performance but also enables interpretable analysis by learning sequence binding motifs and generating attribution maps. Benchmarking diPaRIS across 44 datasets shows its superiority over existing methods. The model consistently achieves the highest accuracy, AUC, AUPR, and F1‐score across all datasets. Additionally, diPaRIS excels in cross‐cell line predictions, consistently outperforming the second‐best method across all datasets. Predictions by diPaRIS reflect the conservation of protein‐RNA binding and facilitate further functional interpretation of genetic variants in complex diseases. The findings highlight that diPaRIS effectively predicts protein‐RNA interactions and interprets potential gene‐disease associations.
- 29 August 2025
- Published by Wiley in Advanced Science
- Vol. 12 (38) , e06314
Open AccessdiPaRIS: Dynamic and Interpretable Protein‐RNA Interactions Prediction With U‐Shaped Network and Novel Structure Encoding
Protein‐RNA interactions play a critical role in various biological processes and disease development. Proteins interact with RNAs through dynamic binding sites that exhibit specific structural patterns under various cellular conditions. While current computational methods take RNA structures in vivo into account, they fall short in capturing the structure contextual association of nucleotides, limiting predictive accuracy. Here, diPaRIS, a deep learning method, is proposed to predict dynamic protein‐RNA interactions with improved accuracy and enhanced interpretability by integrating RNA structures in vivo. diPaRIS introduces a novel encoding scheme for SHAPE‐seq to encode nucleotide correlations, providing a more comprehensive representation of RNA structures. Leveraging a U‐shaped network architecture, diPaRIS not only improves prediction performance but also enables interpretable analysis by learning sequence binding motifs and generating attribution maps. Benchmarking diPaRIS across 44 datasets shows its superiority over existing methods. The model consistently achieves the highest accuracy, AUC, AUPR, and F1‐score across all datasets. Additionally, diPaRIS excels in cross‐cell line predictions, consistently outperforming the second‐best method across all datasets. Predictions by diPaRIS reflect the conservation of protein‐RNA binding and facilitate further functional interpretation of genetic variants in complex diseases. The findings highlight that diPaRIS effectively predicts protein‐RNA interactions and interprets potential gene‐disease associations.
- Open Access
Inference and visualization of multi-scale cell tree for decoding functional diversity with scMustree
: Deciphering cellular heterogeneity and functional diversity is crucial to understanding biological systems and disease mechanisms. Although single-cell RNA sequencing (scRNA-seq) has revolutionized this field, existing methods often fail to characterize the functional associations between cell populations due to the discrete nature of clustering results. We introduce scMustree, a tree- construction algorithm that enables in-depth exploration of functional associations and transitional states among cellular populations. scMustree combines top-down iterative decomposition for high- purity leaf nodes with bottom-up merging based on quantitative functional distances, preserving local continuity and global structural clarity. Benchmarking on real datasets shows that scMustree not only outperforms existing methods in clustering accuracy but also reveals functionally coherent subtypes, anatomically related cell types, and developmentally connected transitions. In case studies, it identifies Alzheimer’s disease-associated microglial subtypes and captures spatial cell distribution patterns of cell types in developed embryos, demonstrating its utility as a powerful tool for multi-scale exploration of cellular heterogeneity.
Open AccessInference and visualization of multi-scale cell tree for decoding functional diversity with scMustree
: Deciphering cellular heterogeneity and functional diversity is crucial to understanding biological systems and disease mechanisms. Although single-cell RNA sequencing (scRNA-seq) has revolutionized this field, existing methods often fail to characterize the functional associations between cell populations due to the discrete nature of clustering results. We introduce scMustree, a tree- construction algorithm that enables in-depth exploration of functional associations and transitional states among cellular populations. scMustree combines top-down iterative decomposition for high- purity leaf nodes with bottom-up merging based on quantitative functional distances, preserving local continuity and global structural clarity. Benchmarking on real datasets shows that scMustree not only outperforms existing methods in clustering accuracy but also reveals functionally coherent subtypes, anatomically related cell types, and developmentally connected transitions. In case studies, it identifies Alzheimer’s disease-associated microglial subtypes and captures spatial cell distribution patterns of cell types in developed embryos, demonstrating its utility as a powerful tool for multi-scale exploration of cellular heterogeneity.