Scholarlink

Abstract <p>Ferroptosis is a distinct iron-dependent form of regulated cell death that plays critical roles in cancer progression, neurodegenerative disorders, and immune regulation. Computational identification of ferroptosis-related proteins (FRPs) remains challenging due to the complex regulatory network of ferroptosis, the functional heterogeneity of FRPs, and the limited availability of experimentally validated data. Accurate and high-throughput prediction of FRPs is therefore urgently needed. To address these challenges, we propose <b>FeroConCap</b>, a novel deep learning framework that integrates fractal chaos game representation (FCGR) encoding, capsule networks, and supervised contrastive learning to capture hierarchical and spatial sequence dependencies associated with ferroptosis. The supervised contrastive strategy enhances intra-class compactness while increasing inter-class separability in the embedding space, leading to more robust and discriminative representations. Using a benchmark dataset of 2298 non-redundant protein sequences, FeroConCap achieved state-of-the-art performance, with an accuracy of 95.65% and an MCC of 0.915, exceeding the current method by 4.13% in accuracy and 0.084 in MCC. Comprehensive ablation studies and feature visualization analyses further confirm that both FCGR encoding and the capsule architecture substantially contribute to performance improvement over traditional handcrafted descriptors. To facilitate practical applications, a user-friendly web server has been developed for efficient and large-scale FRP prediction, freely available at <a href="https://ycclab.cuhk.edu.cn/FeroConCap">https://ycclab.cuhk.edu.cn/FeroConCap</a>.</p> Scientific Contribution <p>This study introduces FeroConCap, a novel deep learning framework that enhances the identification of ferroptosis-related proteins. By integrating FCGR encoding with Capsule Networks and supervised contrastive learning, the method effectively captures complex sequence patterns, outperforming existing methods like FRP-XGBoost. This work provides a user-friendly web server, offering a robust tool for high-throughput screening in ferroptosis research and drug discovery.&#xa0;</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

Contrastive representation learning and capsule networks enable accurate identification of ferroptosis-related proteins

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月28日

查看全文
  • Yiyang Zhao,
  • Xingchen Liu,
  • Peilin Xie,
  • Jiahui Guan,
  • Zhihao Zhao,
  • Junwen Wang,
  • Tzong-Yi Lee,
  • Ying-Chih Chiang,
  • Leyi Wei,
  • Xiangrong Liu,
  • Lantian Yao

摘要

Abstract

Ferroptosis is a distinct iron-dependent form of regulated cell death that plays critical roles in cancer progression, neurodegenerative disorders, and immune regulation. Computational identification of ferroptosis-related proteins (FRPs) remains challenging due to the complex regulatory network of ferroptosis, the functional heterogeneity of FRPs, and the limited availability of experimentally validated data. Accurate and high-throughput prediction of FRPs is therefore urgently needed. To address these challenges, we propose FeroConCap, a novel deep learning framework that integrates fractal chaos game representation (FCGR) encoding, capsule networks, and supervised contrastive learning to capture hierarchical and spatial sequence dependencies associated with ferroptosis. The supervised contrastive strategy enhances intra-class compactness while increasing inter-class separability in the embedding space, leading to more robust and discriminative representations. Using a benchmark dataset of 2298 non-redundant protein sequences, FeroConCap achieved state-of-the-art performance, with an accuracy of 95.65% and an MCC of 0.915, exceeding the current method by 4.13% in accuracy and 0.084 in MCC. Comprehensive ablation studies and feature visualization analyses further confirm that both FCGR encoding and the capsule architecture substantially contribute to performance improvement over traditional handcrafted descriptors. To facilitate practical applications, a user-friendly web server has been developed for efficient and large-scale FRP prediction, freely available at https://ycclab.cuhk.edu.cn/FeroConCap.

Scientific Contribution

This study introduces FeroConCap, a novel deep learning framework that enhances the identification of ferroptosis-related proteins. By integrating FCGR encoding with Capsule Networks and supervised contrastive learning, the method effectively captures complex sequence patterns, outperforming existing methods like FRP-XGBoost. This work provides a user-friendly web server, offering a robust tool for high-throughput screening in ferroptosis research and drug discovery.