Multiple sequence alignment based on deep reinforcement learning with self-attention and positional encoding.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Multiple sequence alignment (MSA) is crucial for sequence analysis but remains challenging due to its NP-complete nature.
• Existing MSA methods have limitations in accuracy and reliability.

Purpose of the Study:

• To develop a more accurate and reliable method for multiple sequence alignment.
• To leverage deep reinforcement learning and natural language processing techniques for MSA.

Main Methods:

• Proposed a deep reinforcement learning framework incorporating positional encoding and self-attention for MSA.
• Utilized positional encoding to preserve nucleotide position information.
• Employed self-attention to extract key sequence features.
• Designed a novel reinforcement learning environment for progressive column alignment.

Main Results:

• The proposed method significantly enhances MSA accuracy.
• Achieved superior performance compared to state-of-the-art methods on benchmark datasets.
• Demonstrated effectiveness using Sum-of-pairs and Column scores.

Conclusions:

• The novel deep reinforcement learning approach offers a promising solution for accurate multiple sequence alignment.
• The integration of NLP techniques like self-attention advances MSA methodology.
• The open-source implementation facilitates further research and application.