Scholarlink

<p>Transcription factor binding sites (TFBS) play pivotal roles in various biological functions, and their precise identification is crucial for deciphering gene regulatory mechanisms. However, conventional experimental approaches are often time-consuming and expensive, while existing computational methods face challenges in further improving prediction accuracy, largely due to their limited ability to capture complex hidden information from DNA sequences. To address these limitations, this study proposes a hybrid deep learning framework named GLNet-TFBS, which integrates dual-path feature learning modules—global and local—to enhance TFBS prediction performance. One path employs the pre-trained DNABERT-2 model, leveraging its Transformer-based architecture to capture rich global contextual information from DNA sequences. The other path incorporates three classical sequence encoding techniques to extract discriminative local patterns. Experimental results show that the fusion of features from both paths significantly improves prediction accuracy and robustness. As a result, our model outperforms existing benchmarks on key evaluation metrics, including ACC, ROC-AUC, and PR-AUC, demonstrating its strong potential for broad applicability across diverse biological contexts. The source code and data are available at&#xa0;<a href="https://github.com/zjxywt/GLNet-TFBS">https://github.com/zjxywt/GLNet-TFBS</a>.</p> Graphical Abstract <p></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

A Deep Learning Framework with Multi-perspective Feature Fusion for Transcription Factor Binding Site Prediction

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Tong Wang,
  • Zhendong Liu

摘要

Transcription factor binding sites (TFBS) play pivotal roles in various biological functions, and their precise identification is crucial for deciphering gene regulatory mechanisms. However, conventional experimental approaches are often time-consuming and expensive, while existing computational methods face challenges in further improving prediction accuracy, largely due to their limited ability to capture complex hidden information from DNA sequences. To address these limitations, this study proposes a hybrid deep learning framework named GLNet-TFBS, which integrates dual-path feature learning modules—global and local—to enhance TFBS prediction performance. One path employs the pre-trained DNABERT-2 model, leveraging its Transformer-based architecture to capture rich global contextual information from DNA sequences. The other path incorporates three classical sequence encoding techniques to extract discriminative local patterns. Experimental results show that the fusion of features from both paths significantly improves prediction accuracy and robustness. As a result, our model outperforms existing benchmarks on key evaluation metrics, including ACC, ROC-AUC, and PR-AUC, demonstrating its strong potential for broad applicability across diverse biological contexts. The source code and data are available at https://github.com/zjxywt/GLNet-TFBS.

Graphical Abstract