基于Conv-Transformer的RNA碱基配对概率预测  

RNA Base Pair Probability Prediction Based on Conv-Transformer

作　　者：李大舟 陈梅 高巍 LI Dazhou;CHEN Mei;GAO Wei(School of Computer Science and Technology,Shenyang University of Chemical Technology,Shenyang 110142)

机构地区：[1]沈阳化工大学计算机科学与技术学院,沈阳110142

出　　处：《计算机与数字工程》2025年第3期781-785,共5页Computer & Digital Engineering

基　　金：辽宁省教育厅科学研究项目(编号:LJKZ0449,LJ2020033)资助。

摘　　要：RNA碱基配对概率衡量了RNA序列在各个点位是否能形成稳定的碱基对,是RNA结构的重要属性,可被用在mRNA疫苗序列设计、疾病诊断、药物研发等领域。论文提出了一种Conv-Transformer模型预测RNA碱基配对概率。论文使用Transformer模型描述RNA碱基的全局特征,使用一维卷积描述RNA碱基的局部特征。论文利用编解码器中的注意力机制提取RNA碱基的全局信息。再将解码器的输出矩阵输入到一维卷积中提取RNA碱基的局部信息。最终由一维卷积输出RNA序列碱基的配对概率。论文采用均方根误差和平均绝对误差作为性能评价指标。实验结果表明,论文提出的Conv-Transformer模型与经典机器学习模型相比拥有更好的表现,均方根误差降低了16%左右,平均绝对误差降低了20%左右。RNA base pairing probability measures whether the RNA sequence can form stable base pairs at various points,which is an important attribute of RNA structure and can be used in mRNA vaccine sequence design,disease diagnosis,drug research and development and other fields.Therefore,a Conv-Transformer model is proposed to predict the probability of RNA base pairing.Transformer model is used to describe the global characteristics of RNA bases and 1D convolution is used to describe the local characteristics of RNA bases.Firstly,the global information of RNA bases is extracted by using Attention.Then,the output matrix of the Decoder is input to the 1D convolution to extract the local information of RNA bases.Finally,the pairing probabilities of RNA sequence bases are output from the 1D convolution.Root mean square error and the average absolute error are used as performance evaluation indicators.Experimental results show that the Conv-Transformer model proposed has better performance than the classical machine learning model,with the root mean square error reduced by about 16%and the average absolute error by about 20%.

关 键 词：RNA碱基配对概率 一维卷积 TRANSFORMER 幂归一化 多头自注意力 RNA结构预测 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]