Scholarlink

<p>Colorectal cancer (CRC) and inflammatory bowel disease (IBD) share complex and overlapping molecular profiles that poses challenges in identifying common hub genes contributing to both conditions. This research proposes Variational Graph Autoencoder-Hub (VGAE-Hub), a graph-based deep learning framework that integrates network topology and latent graph representations using a VGAE to uncover shared hub genes associated with CRC and IBD. Gene expression data for both disease are obtained from the NCBI database to identify overlapping genes. The protein–protein interaction (PPI) network is constructed from the overlapping genes. A comprehensive set of topological features, including degree, closeness, betweenness, eigenvector, PageRank, clustering coefficient and subgraph centrality is extracted from the PPI network. The VGAE model is trained on the normalized adjacency and feature matrices to learn low-dimensional embeddings that capture both structural and attribute information of the network. To evaluate the discriminative capacity of the extracted features, machine learning classifiers are trained on three different feature sets: (i) topological features, (ii) VGAE embeddings and (iii) combined features. Classification experiments demonstrate that the combined feature set consistently outperformed individual feature types, achieving an improvement of <b>+ 5.6% in accuracy</b> and <b>+ 0.07 in ROC-AUC</b> compared to the best single feature set. This highlights the complementary nature of handcrafted and learned features. The Louvain Community detection is applied to VGAE embeddings facilitates the identification of functionally coherent gene modules. Network-level validations, including transcriptional regulation (PDI) and gene regulatory network (GRN) further support the biological relevance of the prioritized hub genes. Additionally, survival analysis and expression validation reveal clinically significant genes like <b>VWF</b>,<b> TIMP1</b>,<b> SPARCL1</b>,<b> CXCL2</b>,<b> SERPINE1</b>,<b> CDH5</b>,<b> FCGR2A</b> and <b>EGR1</b>. Overall, VGAE-Hub offers a scalable and biologically grounded framework for integrative biomarker discovery and provides novel insights into shared molecular mechanisms in CRC and IBD.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

VGAE-Hub: a variational graph autoencoder framework for identifying shared hub genes in colorectal cancer and inflammatory bowel disease

  • 期刊论文

  • 发表日期：

    2025年11月4日

查看全文
  • Pratibha Joshi,
  • Ravi Verma,
  • Buddha Singh

摘要

Colorectal cancer (CRC) and inflammatory bowel disease (IBD) share complex and overlapping molecular profiles that poses challenges in identifying common hub genes contributing to both conditions. This research proposes Variational Graph Autoencoder-Hub (VGAE-Hub), a graph-based deep learning framework that integrates network topology and latent graph representations using a VGAE to uncover shared hub genes associated with CRC and IBD. Gene expression data for both disease are obtained from the NCBI database to identify overlapping genes. The protein–protein interaction (PPI) network is constructed from the overlapping genes. A comprehensive set of topological features, including degree, closeness, betweenness, eigenvector, PageRank, clustering coefficient and subgraph centrality is extracted from the PPI network. The VGAE model is trained on the normalized adjacency and feature matrices to learn low-dimensional embeddings that capture both structural and attribute information of the network. To evaluate the discriminative capacity of the extracted features, machine learning classifiers are trained on three different feature sets: (i) topological features, (ii) VGAE embeddings and (iii) combined features. Classification experiments demonstrate that the combined feature set consistently outperformed individual feature types, achieving an improvement of + 5.6% in accuracy and + 0.07 in ROC-AUC compared to the best single feature set. This highlights the complementary nature of handcrafted and learned features. The Louvain Community detection is applied to VGAE embeddings facilitates the identification of functionally coherent gene modules. Network-level validations, including transcriptional regulation (PDI) and gene regulatory network (GRN) further support the biological relevance of the prioritized hub genes. Additionally, survival analysis and expression validation reveal clinically significant genes like VWF, TIMP1, SPARCL1, CXCL2, SERPINE1, CDH5, FCGR2A and EGR1. Overall, VGAE-Hub offers a scalable and biologically grounded framework for integrative biomarker discovery and provides novel insights into shared molecular mechanisms in CRC and IBD.