Scholarlink

<p>Extensive research has shown that microbial communities are closely linked to the development of many complex human diseases. Therefore, identifying potential microbe-disease associations is essential for improving diagnosis, prognosis, and treatment strategies. Traditional biomedical experiments, however, tend to be costly, time-consuming, and labor-intensive. To address these limitations, we propose a novel computational model, CFAESVD, for predicting potential microbe-disease associations. The model begins by integrating data from three different databases and employs Centered Kernel Alignment-based Multi-Kernel Learning (CKA-MKL) algorithm to fuse multi-source features of microbes and diseases based on known associations. It then utilizes three distinct modules to extract feature representations: four-layer autoencoder to capture the nonlinear features of diseases, Singular Value Decomposition (SVD) module to extract linear features for both microbes and diseases, and Graph Attention Autoncoder (GATE) module for the nonlinear features of microbes. These features are concatenated into feature vectors representing microbe-disease pairs. Finally, these vectors are input into a modified Cascade Forest model for prediction. Experimental results show that under five-fold cross-validation, the model achieves an AUC of 0.9748 and an AUPR of 0.9732, surpassing several advanced prediction models. Additionally, case studies confirm the CFAESVD model’s effectiveness in predicting microbial associations with conditions such as obesity and Crohn’s disease, further validating its reliability. CFAESVD is publicly available at <a href="https://github.com/senliyang/CFAESVD">https://github.com/senliyang/CFAESVD</a>.</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

Predicting microbe-disease association based on multi-kernel autoencoder and singular value decomposition

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Xiaoxin Du,
  • Lisen Yang,
  • Bo Wang,
  • Mei Jin,
  • Zhenfei Wang,
  • Jianfei Zhang

摘要

Extensive research has shown that microbial communities are closely linked to the development of many complex human diseases. Therefore, identifying potential microbe-disease associations is essential for improving diagnosis, prognosis, and treatment strategies. Traditional biomedical experiments, however, tend to be costly, time-consuming, and labor-intensive. To address these limitations, we propose a novel computational model, CFAESVD, for predicting potential microbe-disease associations. The model begins by integrating data from three different databases and employs Centered Kernel Alignment-based Multi-Kernel Learning (CKA-MKL) algorithm to fuse multi-source features of microbes and diseases based on known associations. It then utilizes three distinct modules to extract feature representations: four-layer autoencoder to capture the nonlinear features of diseases, Singular Value Decomposition (SVD) module to extract linear features for both microbes and diseases, and Graph Attention Autoncoder (GATE) module for the nonlinear features of microbes. These features are concatenated into feature vectors representing microbe-disease pairs. Finally, these vectors are input into a modified Cascade Forest model for prediction. Experimental results show that under five-fold cross-validation, the model achieves an AUC of 0.9748 and an AUPR of 0.9732, surpassing several advanced prediction models. Additionally, case studies confirm the CFAESVD model’s effectiveness in predicting microbial associations with conditions such as obesity and Crohn’s disease, further validating its reliability. CFAESVD is publicly available at https://github.com/senliyang/CFAESVD.