Metagenomic sequence classification based on local sensitive hashing and Bi-LSTM

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Metagenomic classification methods often struggle with taxonomic resolution due to short k-mer lengths and reliance on extensive databases.
• Existing techniques can be computationally intensive and limited in their ability to accurately identify species and genera within complex microbial communities.

Purpose of the Study:

• To develop an advanced metagenomic sequence classification method that overcomes the limitations of current approaches.
• To improve taxonomic resolution at the genus and species levels.
• To reduce the dependency on large reference databases in metagenomic analysis.

Main Methods:

• Integration of Locality-Sensitive Hashing (LSH) for k-mer representation.
• Application of the skip-gram model for k-mer embedding.
• Utilizing Bidirectional Long-Short Term Memory (Bi-LSTM) networks for sequence classification.

Main Results:

• The proposed method demonstrates superior classification performance at the genus level compared to existing models.
• The approach effectively learns discriminative features directly from sequences, supporting longer k-mers.
• Reduced runtime reliance on reference databases was achieved.

Conclusions:

• The LSH and Bi-LSTM integrated method offers a significant advancement in metagenomic classification accuracy and efficiency.
• This novel approach provides a powerful tool for analyzing complex metagenomic data.
• Future applications may include rapid clinical pathogen detection.