Scholarlink

<p>With emerging single-cell transcriptomics data, deep learning approaches have enabled the diagnosis of neurodegenerative disorders such as Parkinson’s disease(PD). Based on whole-blood RNA sequencing data, the graph neural network has predicted Parkinson’s disease(PD) progression trajectories by exploiting spatial and temporal views that were related with neurodegenerative disorders. The single-view learning scheme, which focuses on gene expression embedding, was still insufficient in analyzing complicated human brain diseases. As a type of spatial representation underlying transcriptomics data, gene graphs that are closely associated with disease states have been inferred to investigate regulatory mechanisms and molecular dynamics. For disease-specific gene graphs, global and local structures contribute to fine-grained spatial representations. With the dynamic graph attention network (DGAT) backbone, this study proposes a multi-view, multi-scale M<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\vphantom{0}^2\)</EquationSource> </InlineEquation>DGAT method to predict disease progression trajectories for neurodegenerative disorders. Temporal and spatial views have been integrated by the count sketch bilinear (CSB) fusion strategy. Based on RNA sequencing data from human blood samples, joint-view representations have been constructed to predict disease stages and relevant cognitive scores. Experiments about the PPMI and PDBP cohorts have validated the effectiveness and efficiency of the proposed M<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\vphantom{0}^2\)</EquationSource> </InlineEquation>DGAT architecture in disease prediction applications. The proposed M<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(\vphantom{0}^2\)</EquationSource> </InlineEquation>DGAT method has demonstrated significant superiority in predictive accuracy over established cutting-edge disease prediction approaches. Compared with static graphs, dynamic graph representations tend to encode more dynamics about disease progression.</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

M\(\vphantom{0}^2\)DGAT: Multi-view multi-scale dynamic graph attention network(GAT) based prediction of Parkinson’s disease(PD) progression using whole-blood RNA sequencing data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月2日

查看全文
  • Zhang Wei,
  • Xu Zeqi,
  • Wu Chenjun,
  • Dai Qi

摘要

With emerging single-cell transcriptomics data, deep learning approaches have enabled the diagnosis of neurodegenerative disorders such as Parkinson’s disease(PD). Based on whole-blood RNA sequencing data, the graph neural network has predicted Parkinson’s disease(PD) progression trajectories by exploiting spatial and temporal views that were related with neurodegenerative disorders. The single-view learning scheme, which focuses on gene expression embedding, was still insufficient in analyzing complicated human brain diseases. As a type of spatial representation underlying transcriptomics data, gene graphs that are closely associated with disease states have been inferred to investigate regulatory mechanisms and molecular dynamics. For disease-specific gene graphs, global and local structures contribute to fine-grained spatial representations. With the dynamic graph attention network (DGAT) backbone, this study proposes a multi-view, multi-scale M \(\vphantom{0}^2\) DGAT method to predict disease progression trajectories for neurodegenerative disorders. Temporal and spatial views have been integrated by the count sketch bilinear (CSB) fusion strategy. Based on RNA sequencing data from human blood samples, joint-view representations have been constructed to predict disease stages and relevant cognitive scores. Experiments about the PPMI and PDBP cohorts have validated the effectiveness and efficiency of the proposed M \(\vphantom{0}^2\) DGAT architecture in disease prediction applications. The proposed M \(\vphantom{0}^2\) DGAT method has demonstrated significant superiority in predictive accuracy over established cutting-edge disease prediction approaches. Compared with static graphs, dynamic graph representations tend to encode more dynamics about disease progression.