Scholarlink

<p>Protein function prediction, protein–protein interaction (PPI) prediction, and complex identification are essential tasks in bioinformatics and play a critical role in understanding biological mechanisms underlying food traits such as stress resistance, nutritional content, and resilience. This is especially relevant for endangered and underutilized edible species, where limited molecular characterization hinders conservation and sustainable use. Due to the scale-free and hierarchical nature of PPI networks—often dominated by hub proteins—traditional Euclidean space embedding methods fail to capture their structural complexity, leading to suboptimal protein representations. In this study, we propose a protein autoencoder based on a hyperbolic space graph embedding model (HVGA), designed to effectively encode hierarchical relationships within PPI networks. The model employs two hyperbolic graph convolutional networks (HGCNs) as encoders to compute the mean and variance of the hidden layer, capturing network topology across varying curvatures. A Fermi-Dirac decoder reconstructs the PPI network through hyperbolic inner product operations. Experimental evaluations demonstrate that our model significantly outperforms Euclidean-based approaches in downstream tasks, achieving approximately 0.07 higher AUC values in PPI prediction and 0.02 higher Macro-F1 scores in protein function prediction on three benchmark PPI datasets. The results support HVGA’s potential for extracting functional protein insights in endangered food species, contributing to biodiversity preservation, climate-resilient agriculture, and sustainable food system development.</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

Hyperbolic graph autoencoder-based functional protein network analysis for preserving edible biodiversity in endangered food species

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月15日

查看全文
  • Haewon Byeon,
  • Ajith Sundaram,
  • Komatigunta Nagaraju,
  • Elangovan Muniyandy,
  • Mukesh Soni,
  • Mohammed Wasim Bhatt,
  • Mulugeta Tesema

摘要

Protein function prediction, protein–protein interaction (PPI) prediction, and complex identification are essential tasks in bioinformatics and play a critical role in understanding biological mechanisms underlying food traits such as stress resistance, nutritional content, and resilience. This is especially relevant for endangered and underutilized edible species, where limited molecular characterization hinders conservation and sustainable use. Due to the scale-free and hierarchical nature of PPI networks—often dominated by hub proteins—traditional Euclidean space embedding methods fail to capture their structural complexity, leading to suboptimal protein representations. In this study, we propose a protein autoencoder based on a hyperbolic space graph embedding model (HVGA), designed to effectively encode hierarchical relationships within PPI networks. The model employs two hyperbolic graph convolutional networks (HGCNs) as encoders to compute the mean and variance of the hidden layer, capturing network topology across varying curvatures. A Fermi-Dirac decoder reconstructs the PPI network through hyperbolic inner product operations. Experimental evaluations demonstrate that our model significantly outperforms Euclidean-based approaches in downstream tasks, achieving approximately 0.07 higher AUC values in PPI prediction and 0.02 higher Macro-F1 scores in protein function prediction on three benchmark PPI datasets. The results support HVGA’s potential for extracting functional protein insights in endangered food species, contributing to biodiversity preservation, climate-resilient agriculture, and sustainable food system development.