Pairwise Heuristic Sequence Alignment Algorithm Based on Deep Reinforcement Learning

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Sequence alignment is crucial for comparative genomic analysis.
• Existing pairwise alignment algorithms face limitations, particularly with dissimilar sequences.
• Deep reinforcement learning offers a novel approach to enhance sequence alignment.

Purpose of the Study:

• To introduce a new pairwise sequence alignment method utilizing deep reinforcement learning.
• To address limitations of conventional alignment algorithms.
• To improve the analysis of genomic sequence associations.

Main Methods:

• Development of a deep reinforcement learning framework for sequence alignment.
• Definition of environment and agent for the reinforcement learning system.
• Implementation of DQNalign, which uses a moving window for subsequence observation.

Main Results:

• DQNalign demonstrates superior performance in aligning dissimilar sequence pairs with low identity.
• Theoretical analysis confirms DQNalign's low-dimensional complexity relative to sequence length.
• The method effectively determines alignment directions based on observed subsequences.

Conclusions:

• Deep reinforcement learning can be successfully applied to sequence alignment systems.
• DQNalign represents a significant improvement over conventional pairwise sequence alignment methods.
• This research highlights the potential of AI in advancing genomic sequence analysis.