Scholarlink

<p>Promoters are crucial parts of DNA sequences and play a significant role in understanding the process of gene regulation. Despite the existence of numerous methods for predicting promoters, they are typically built on single-sequence information and seldom incorporate external knowledge graphs (KGs). Knowledge graphs, however, can provide rich factual knowledge that enhances the accuracy of promoter identification. In this study, we propose a novel method called OntoGene, which integrates Gene Ontology (GO) into the bidirectional encoder representations from transformers model (BERT) for promoter identification. Specifically, OntoGene encodes both gene sequences and external knowledge into vectors using a BERT encoder. Our method jointly optimizes knowledge embedding and masked language modeling objectives, and utilizes a contrastive learning method with knowledge-aware negative sampling to optimize gene and knowledge graph embeddings. OntoGene is a BERT-based model designed to predict not only the presence of promoters but also their strength (strong or weak promoters) from gene sequences. Experimental results demonstrate OntoGene’s high accuracy in predicting DNA promoters and their strengths, validating the effectiveness of this knowledge-enhanced approach. OntoGene and the dataset can be obtained from <a href="https://github.com/gaomengli7/OntoGene">https://github.com/gaomengli7/OntoGene</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

OntoGene: knowledge-enhanced BERT for promoter identification

  • 期刊论文

  • 发表日期：

    2025年6月23日

查看全文
  • Yang Li,
  • Mengli Gao,
  • Jilong Bian,
  • Kaiqi Zhao,
  • Dan Li,
  • Guohua Wang

摘要

Promoters are crucial parts of DNA sequences and play a significant role in understanding the process of gene regulation. Despite the existence of numerous methods for predicting promoters, they are typically built on single-sequence information and seldom incorporate external knowledge graphs (KGs). Knowledge graphs, however, can provide rich factual knowledge that enhances the accuracy of promoter identification. In this study, we propose a novel method called OntoGene, which integrates Gene Ontology (GO) into the bidirectional encoder representations from transformers model (BERT) for promoter identification. Specifically, OntoGene encodes both gene sequences and external knowledge into vectors using a BERT encoder. Our method jointly optimizes knowledge embedding and masked language modeling objectives, and utilizes a contrastive learning method with knowledge-aware negative sampling to optimize gene and knowledge graph embeddings. OntoGene is a BERT-based model designed to predict not only the presence of promoters but also their strength (strong or weak promoters) from gene sequences. Experimental results demonstrate OntoGene’s high accuracy in predicting DNA promoters and their strengths, validating the effectiveness of this knowledge-enhanced approach. OntoGene and the dataset can be obtained from https://github.com/gaomengli7/OntoGene.