Alignment-free comparison of metagenomics sequences via approximate string matching

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Accurate pairwise sequence similarity quantification is crucial for metagenomics.
• Alignment-free methods offer computational efficiency for large-scale sequence analysis.
• Existing neural network methods struggle with variable sequence lengths and indels.

Purpose of the Study:

• To develop a novel alignment-free method, AsMac, for robust sequence similarity quantification.
• To address limitations of current methods regarding sequence length variation and insertions/deletions.
• To provide an efficient and effective tool for metagenomic data analysis.

Main Methods:

• Proposed a novel neural network architecture for approximate string matching.
• Developed an efficient gradient computation algorithm for neural network training.
• Utilized real-world data for large-scale benchmarking.

Main Results:

• The AsMac method demonstrates effectiveness in quantifying pairwise sequence similarities.
• The approach successfully handles sequences of varying lengths.
• The method shows improved performance in the presence of insertions and deletions.

Conclusions:

• AsMac offers a significant advancement in alignment-free sequence analysis for metagenomics.
• The developed method is computationally efficient and robust.
• Open-source software and trained models are available for broader application.