Using deep reinforcement learning approach for solving the multiple sequence alignment problem

In the present paper, we use a deep reinforcement learning (DRL) approach for solving the multiple sequence alignment problem which is an NP-complete problem. Multiple Sequence Alignment problem simply refers to the process of arranging initial sequences of DNA, RNA or proteins in order to maximize their regions of similarity. Multiple Sequence Alignment is the first step in solving many bioinformatics problems such as constructing phylogenetic trees. In this study, our proposed approach models the Multiple Sequence Alignment problem as a DRL problem and utilizes long short-term memory networks for estimation phase in the reinforcement learning algorithm. Furthermore, the actor-critic algorithm with experience-replay method is used for much quicker convergence process. Using deep Q-learning (an RL approach) and Q-network overcomes the complexity of other approaches. The experimental evaluation is performed on 8 different real-life datasets and in every used dataset our approach outperforms other well-known approaches and tools such as MAFFT, ClustalW, and other heuristic approaches in case of scoring in solving the MSA problem.