基于强化学习的双生物序列比对优化算法  

The Optimization Algorithm for Pairwise Biological Sequence Alignment Based on Reinforcement Learning

作　　者：石海鹤[1] 江浪 陈名森 王岚[1] SHI Haihe;JIANG Lang;CHEN Mingsen;WANG Lan(College of Computer Information Engineering,Jiangxi Normal University,Nanchang Jiangxi 330022,China)

机构地区：[1]江西师范大学计算机信息工程学院,江西南昌330022

出　　处：《江西师范大学学报(自然科学版)》2024年第5期464-471,490,共9页Journal of Jiangxi Normal University(Natural Science Edition)

基　　金：国家自然科学基金(62062039)资助项目.

摘　　要：该文设计了一种基于强化学习的双生物序列比对算法—QLalign,构建了序列比对问题的序列决策环境模型.基于Q-Learning算法的核心思想,将QLalign与环境模型进行动作决策-信息反馈的交互训练,不断优化决策的正确率,得到双生物序列比对问题的最优解.与Needleman-Wunsch算法进行了比较实验,验证了QLalign算法的可行性,而且对比结果表明:在相似度高于60%的序列对上,QLalign算法较经典的动态规划算法在效率方面展示出极大的优越性.此外,在QLalign算法训练过程中所记录的次优解可作为序列最优比对候选集,供进一步分析和研究.The pairwise biological sequence alignment algorithm QLalign based on reinforcement learning is designed,and the sequence decision environment model for sequence alignment problem is constructed.Based on the core idea of Q-learning algorithm,QLalign and the environment model carry out the interactive training of action decision-making and information feedback.The feedback information is continuously optimized through calculation of the correct rate of decision making,and the optimal solution of biological sequence alignment problem is obtained through the final learned optimal decision.Compared with Needleman-Wunsch algorithm,the feasibility of QLalign algorithm is verified.The results show that QLalign algorithm has great efficiency advantages compared with classical dynamic programming algorithm for sequence pairs with similarity higher than 60%.In addition,the sub-optimal solutions recorded during the training process of QLalign algorithm can be used as the candidate set of optimal sequence alignment for further analysis and research.

关 键 词：生物信息学 序列比对 强化学习 Q-LEARNING 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]