Wenju Hou
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Wenju Hou has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Yan Wang
Jilin University
6 shared publications Chunman Zuo
2 shared publications- Jie Hong
College of Computer Science and Technology, Jilin University, Changchun, China
2 shared publications - Nan Sheng
Key Laboratory of Symbol Computation and Knowledge Engineering of Ministry of Education, College of Computer Science and Technology, Jilin University, Changchun, China
2 shared publications - Wei Pang
School of Mathematical and Computer Sciences, Heriot-Watt University, Edinburgh, Scotland, United Kingdom
2 shared publications - Yizhi Cong
School of Artificial Intelligence, Jilin University, Changchun, China
2 shared publications
Latest Publications
- 17 October 2026
- Published by Elsevier in Pattern Recognition
- Vol. 183
Rebuilding to remember: Reconstruction-driven regularization for information-preserving in clustering graph pooling
- 17 October 2026
- Published by Elsevier in Pattern Recognition
- Vol. 183
Rebuilding to remember: Reconstruction-driven regularization for information-preserving in clustering graph pooling
- 25 September 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 260
HATZFS predicts pancreatic cancer driver biomarkers by hierarchical reinforcement learning and zero-forcing set
- 25 September 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 260
HATZFS predicts pancreatic cancer driver biomarkers by hierarchical reinforcement learning and zero-forcing set
- 24 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 29 (1) , 657-668
Self-Supervised Contrastive Learning on Attribute and Topology Graphs for Predicting Relationships Among lncRNAs, miRNAs and Diseases
Exploring potential association between long non-coding RNAs (lncRNAs), microRNAs (miRNAs) and diseases is an essential part of prevention, diagnosis and treatment of diseases. Since determining these relationships experimentally is resource-intensive and time-consuming, therefore computational methods have emerged as an attractive way to address this issue. However, existing computational approaches for inferring lncRNA-disease associations (LDA), miRNA-disease associations (MDA) and lncRNA-miRNA interactions (LMI) tend to focus on single task, neglecting the benefits of leveraging multiple biomolecular interactions and domain-specific knowledge for multi-task prediction. Furthermore, labeled data for LDA, MDA and LMI is scarce and costly in real-word applications, making it challenging for models to learn comprehensive node embedding patterns. Building on our previous work, this paper proposes a multi-task prediction model (called SSCLMD) that employs self-supervised contrastive learning on attribute and topology graphs to identify potential LDAs, MDAs and LMIs. Specifically, firstly, domain knowledge of lncRNAs, miRNAs and diseases as well as their interactions are exploited to construct attribute graph and topology graph, respectively. Then, the nodes are encoded in the attribute and topology spaces to extract the specific and common feature. Meanwhile, the attention mechanism is performed to adaptively fuse the embedding from different views. SSCLMD incorporates a contrastive self-supervised learning task as a regularize to guide the learning of node embeddings in both attribute and topology space without relying on labels. Severing as a regularize in multi-task learning paradigm, it to improves the model's generalization capabilities. Extensive experiments on 2 manually curated datasets demonstrate that SSCLMD significantly outperforms other baseline methods in LDA, MDA and LMI prediction tasks. Additionally, case studies on both new and old datasets further supported the ability of SSCLMD to uncover novel disease-related lncRNAs and miRNAs. The source codes and supplementary file of this work are publicly available on \url{https://github.com/sheng-n/SSCLMD}.
- 24 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 29 (1) , 657-668
Self-Supervised Contrastive Learning on Attribute and Topology Graphs for Predicting Relationships Among lncRNAs, miRNAs and Diseases
Exploring potential association between long non-coding RNAs (lncRNAs), microRNAs (miRNAs) and diseases is an essential part of prevention, diagnosis and treatment of diseases. Since determining these relationships experimentally is resource-intensive and time-consuming, therefore computational methods have emerged as an attractive way to address this issue. However, existing computational approaches for inferring lncRNA-disease associations (LDA), miRNA-disease associations (MDA) and lncRNA-miRNA interactions (LMI) tend to focus on single task, neglecting the benefits of leveraging multiple biomolecular interactions and domain-specific knowledge for multi-task prediction. Furthermore, labeled data for LDA, MDA and LMI is scarce and costly in real-word applications, making it challenging for models to learn comprehensive node embedding patterns. Building on our previous work, this paper proposes a multi-task prediction model (called SSCLMD) that employs self-supervised contrastive learning on attribute and topology graphs to identify potential LDAs, MDAs and LMIs. Specifically, firstly, domain knowledge of lncRNAs, miRNAs and diseases as well as their interactions are exploited to construct attribute graph and topology graph, respectively. Then, the nodes are encoded in the attribute and topology spaces to extract the specific and common feature. Meanwhile, the attention mechanism is performed to adaptively fuse the embedding from different views. SSCLMD incorporates a contrastive self-supervised learning task as a regularize to guide the learning of node embeddings in both attribute and topology space without relying on labels. Severing as a regularize in multi-task learning paradigm, it to improves the model's generalization capabilities. Extensive experiments on 2 manually curated datasets demonstrate that SSCLMD significantly outperforms other baseline methods in LDA, MDA and LMI prediction tasks. Additionally, case studies on both new and old datasets further supported the ability of SSCLMD to uncover novel disease-related lncRNAs and miRNAs. The source codes and supplementary file of this work are publicly available on \url{https://github.com/sheng-n/SSCLMD}.
- Open Access
- 17 July 2023
- Published by Springer Nature in Complex & Intelligent Systems
- Vol. 10 (1) , 111-128
Hierarchical graph neural network with subgraph perturbations for key gene cluster discovery in cancer staging
Analyzing highly individual-specific genomic data to understand genetic interactions in cancer development is still challenging, with significant implications for the discovery of individual biomarkers as well as personalized medicine. With the rapid development of deep learning, graph neural networks (GNNs) have been employed to analyze a wide range of biomolecular networks. However, many neural networks are limited to black box models, which are only capable of making predictions, and they are often challenged to provide reliable biological and clinical insights. In this research, for sample-specific networks, a novel end-to-end hierarchical graph neural network with interpretable modules is proposed, which learns structural features at multiple scales and incorporates a soft mask layer in extracting subgraphs that contribute to classification. The perturbations caused by the input graphs' deductions are used to evaluate key gene clusters, and the samples are then grouped into classes to produce both sample- and stage-level explanations. Experiments on four gene expression datasets from The Cancer Genome Atlas (TCGA) show that the proposed model not only rivals the advanced GNN methods in cancer staging but also identifies key gene clusters that have a great impact on classification confidence, providing potential targets for personalized medicine.
- 17 July 2023
- Published by Springer Nature in Complex & Intelligent Systems
- Vol. 10 (1) , 111-128
Open AccessHierarchical graph neural network with subgraph perturbations for key gene cluster discovery in cancer staging
Analyzing highly individual-specific genomic data to understand genetic interactions in cancer development is still challenging, with significant implications for the discovery of individual biomarkers as well as personalized medicine. With the rapid development of deep learning, graph neural networks (GNNs) have been employed to analyze a wide range of biomolecular networks. However, many neural networks are limited to black box models, which are only capable of making predictions, and they are often challenged to provide reliable biological and clinical insights. In this research, for sample-specific networks, a novel end-to-end hierarchical graph neural network with interpretable modules is proposed, which learns structural features at multiple scales and incorporates a soft mask layer in extracting subgraphs that contribute to classification. The perturbations caused by the input graphs' deductions are used to evaluate key gene clusters, and the samples are then grouped into classes to produce both sample- and stage-level explanations. Experiments on four gene expression datasets from The Cancer Genome Atlas (TCGA) show that the proposed model not only rivals the advanced GNN methods in cancer staging but also identifies key gene clusters that have a great impact on classification confidence, providing potential targets for personalized medicine.
- Open Access
- 10 February 2023
- Published by MDPI AG in Brain Sciences
- Vol. 13 (2) , 306
Dual Semi-Supervised Learning for Classification of Alzheimer’s Disease and Mild Cognitive Impairment Based on Neuropsychological Data
Deep learning has shown impressive diagnostic abilities in Alzheimer’s disease (AD) research in recent years. However, although neuropsychological tests play a crucial role in screening AD and mild cognitive impairment (MCI), there is still a lack of deep learning algorithms only using such basic diagnostic methods. This paper proposes a novel semi-supervised method using neuropsychological test scores and scarce labeled data, which introduces difference regularization and consistency regularization with pseudo-labeling. A total of 188 AD, 402 MCI, and 229 normal controls (NC) were enrolled in the study from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. We first chose the 15 features most associated with the diagnostic outcome by feature selection among the seven neuropsychological tests. Next, we proposed a dual semi-supervised learning (DSSL) framework that uses two encoders to learn two different feature vectors. The diagnosed 60 and 120 subjects were randomly selected as training labels for the model. The experimental results show that DSSL achieves the best accuracy and stability in classifying AD, MCI, and NC (85.47% accuracy for 60 labels and 88.40% accuracy for 120 labels) compared to other semi-supervised methods. DSSL is an excellent semi-supervised method to provide clinical insight for physicians to diagnose AD and MCI.
- 10 February 2023
- Published by MDPI AG in Brain Sciences
- Vol. 13 (2) , 306
Open AccessDual Semi-Supervised Learning for Classification of Alzheimer’s Disease and Mild Cognitive Impairment Based on Neuropsychological Data
Deep learning has shown impressive diagnostic abilities in Alzheimer’s disease (AD) research in recent years. However, although neuropsychological tests play a crucial role in screening AD and mild cognitive impairment (MCI), there is still a lack of deep learning algorithms only using such basic diagnostic methods. This paper proposes a novel semi-supervised method using neuropsychological test scores and scarce labeled data, which introduces difference regularization and consistency regularization with pseudo-labeling. A total of 188 AD, 402 MCI, and 229 normal controls (NC) were enrolled in the study from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. We first chose the 15 features most associated with the diagnostic outcome by feature selection among the seven neuropsychological tests. Next, we proposed a dual semi-supervised learning (DSSL) framework that uses two encoders to learn two different feature vectors. The diagnosed 60 and 120 subjects were randomly selected as training labels for the model. The experimental results show that DSSL achieves the best accuracy and stability in classifying AD, MCI, and NC (85.47% accuracy for 60 labels and 88.40% accuracy for 120 labels) compared to other semi-supervised methods. DSSL is an excellent semi-supervised method to provide clinical insight for physicians to diagnose AD and MCI.
- Open Access
Feature Correspondences Increase and Hybrid Terms Optimization Warp for Image Stitching
Feature detection and correct matching are the basis of the image stitching process. Whether the matching is correct and the number of matches directly affect the quality of the final stitching results. At present, almost all image stitching methods use SIFT+RANSAC pattern to extract and match feature points. However, it is difficult to obtain sufficient correct matching points in low-textured or repetitively-textured regions, resulting in insufficient matching points in the overlapping region, and this further leads to the warping model being estimated erroneously. In this paper, we propose a novel and flexible approach by increasing feature correspondences and optimizing hybrid terms. It can obtain sufficient correct feature correspondences in the overlapping region with low-textured or repetitively-textured areas to eliminate misalignment. When a weak texture and large parallax coexist in the overlapping region, the alignment and distortion often restrict each other and are difficult to balance. Accurate alignment is often accompanied by projection distortion and perspective distortion. Regarding this, we propose hybrid terms optimization warp, which combines global similarity transformations on the basis of initial global homography and estimates the optimal warping by adjusting various term parameters. By doing this, we can mitigate projection distortion and perspective distortion, while effectively balancing alignment and distortion. The experimental results demonstrate that the proposed method outperforms the state-of-the-art in accurate alignment on images with low-textured areas in the overlapping region, and the stitching results have less perspective and projection distortion.
Open AccessFeature Correspondences Increase and Hybrid Terms Optimization Warp for Image Stitching
Feature detection and correct matching are the basis of the image stitching process. Whether the matching is correct and the number of matches directly affect the quality of the final stitching results. At present, almost all image stitching methods use SIFT+RANSAC pattern to extract and match feature points. However, it is difficult to obtain sufficient correct matching points in low-textured or repetitively-textured regions, resulting in insufficient matching points in the overlapping region, and this further leads to the warping model being estimated erroneously. In this paper, we propose a novel and flexible approach by increasing feature correspondences and optimizing hybrid terms. It can obtain sufficient correct feature correspondences in the overlapping region with low-textured or repetitively-textured areas to eliminate misalignment. When a weak texture and large parallax coexist in the overlapping region, the alignment and distortion often restrict each other and are difficult to balance. Accurate alignment is often accompanied by projection distortion and perspective distortion. Regarding this, we propose hybrid terms optimization warp, which combines global similarity transformations on the basis of initial global homography and estimates the optimal warping by adjusting various term parameters. By doing this, we can mitigate projection distortion and perspective distortion, while effectively balancing alignment and distortion. The experimental results demonstrate that the proposed method outperforms the state-of-the-art in accurate alignment on images with low-textured areas in the overlapping region, and the stitching results have less perspective and projection distortion.