一个新的强化学习多序列对比工具CDRL  

A New Reinforcement Learning Multi Sequence Comparison Tool-CDRL Model

作　　者：王韦添 江育娥 WANG Weitian;JIANG Yu'e(College of Computer and Cyber Security,Fujian Normal University,Fuzhou 350117,China)

机构地区：[1]福建师范大学计算机与网络空间安全学院,福建福州350117

出　　处：《福建师范大学学报（自然科学版）》2023年第6期40-51,共12页Journal of Fujian Normal University：Natural Science Edition

基　　金：国家自然科学基金资助项目(61472082)。

摘　　要：多序列比对(multiple sequence alignment, MSA)在生物信息学中是一项重要的研究领域,常被用于描述物种之间的进化关系、药物设计和药物开发.MSA是一个NP完全问题,因计算过于复杂,无法获得最优解.强化学习方法在MSA中表现出了优异的性能,但其计算复杂度与空间复杂度都很高,因此无法应用于大规模数据集.提出一种新的强化学习模型CDRL(contextual deep reinforcement learning)来解决多序列比对问题,该模型采用上下文关系,将网络输入维度从O(n2)降为O(n),其中n是输入的序列数量.该模型建立的网络收敛速度快于当前其他模型.实验结果表明,CDRL的性能优于业内其他强化学习MSA.相较于其他方法目前只能运行在12条序列数据上,CDRL成功地在100条序列上取得较快速度和较好性能.这提高了强化学习MSA应用在较大规模数据集上的可能性.Multiple sequence alignment(MSA)is an important research field in bioinformat-ics,which is commonly used to describe the evolutionary relationship between species,and for applications in drug design and drug development.MSA is an NP-complete problem that is too computationally complex to be optimized completely.While reinforcement learning methods have shown excellent performance in MSA,their computational complexity and space complexity hinder their application to large-scale data sets.This paper proposes a new reinforcement learning model,CDRL(Contextual Deep Reinforcement Learning),to solve the problem of multiple sequence alignment.CDRL utilizes contextual relationships to reduce the learning network's parameters from O(n²)to O(n),where n represents the number of input sequences.The network convergence speed established by this model surpasses that of the other state-of-the-art models.Experimental results indicate that CDRL outperforms other MSA methods in the field.Unlike other state-of-the-art methods limited to 12 sequences,CDRL is successfully applied to 100 sequences with better performance and faster running time.This study enhances the possibility of reinforcement MSA learning methods to handle larger data sets.

关 键 词：多序列比对 强化学习 较大规模数据集 空间复杂度 上下文关系 

分 类 号：O172.2[理学—数学]