基于DNABERT编码的启动子预测模型  

A Promoter Prediction Model Based on DNABERT Encoding

作　　者：李伟豪 刘喆 林关宁 LI Weihao;LIU Zhe;LIN Guanning(School of Biomedical Engineering,Shanghai Jiao Tong University,Shanghai 200030,China)

机构地区：[1]上海交通大学生物医学工程学院,上海200030

出　　处：《生物医学工程学进展》2023年第4期364-370,共7页Progress in Biomedical Engineering

基　　金：国家自然科学基金(82150610506)。

摘　　要：启动子是位于基因上游区域的特定DNA序列,通过识别和预测DNA序列中的启动子,可以更好地理解基因调控的机制,促进生物学和医学研究的进展。通过实验的方法来预测启动子既昂贵又费时,而通过计算方法进行启动子预测同样存在不足之处,如精度有待提升、序列编码方式所包含的信息量不足等。该文提出了一种新的编码方式,将预训练模型DNABERT应用于启动子预测的编码,并测试了使用不同深度学习模型进行预测的效果。实验结果表明,使用经过预训练和微调的DNABERT进行编码的Transformer模型在启动子预测任务中取得了较好的效果。Promoters are specific DNA sequences located in the upstream region of genes.By identifying and predicting promoters in DNA sequences,a better understanding of gene regulation mechanisms can be achieved,thereby advancing biological and medical research.Experimental methods for promoter prediction are both expensive and time-consuming.Additionally,computational methods for promoter prediction have limitations such as room for improvement in accuracy and insufficient information content in sequence encoding methods.This paper introduces a novel encoding approach that applies the pre-trained model DNABERT to promoter prediction.Different deep learning models are tested for prediction performance.Experimental results demonstrate that the Transformer model encoded using pre-trained and fine-tuned DNABERT achieves the best performance in promoter prediction tasks.

关 键 词：启动子预测 DNA编码 深度学习 

分 类 号：Q811.4[生物学—生物工程]