Kmer-Node2Vec: a Fast and Efficient Method for Kmer Embedding from the Kmer Co-occurrence Graph, with Applications to...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Learning low-dimensional vector representations of short k-mers from DNA sequences is crucial for bioinformatics tasks.
• Existing methods like DNA2Vec face scalability challenges due to long training times for k-mer embedding.
• Efficient DNA sequence modeling is essential for applications like sequence retrieval and classification.

Purpose of the Study:

• To propose an efficient graph-based k-mer embedding method, Kmer-Node2Vec.
• To address the scalability and training time limitations of existing DNA sequence embedding techniques.
• To achieve fast and high-quality k-mer embeddings for improved DNA sequence modeling.

Main Methods:

• Developed Kmer-Node2Vec, a novel graph-based approach for k-mer embedding.
• Constructed a k-mer co-occurrence graph from large DNA datasets.
• Utilized random walks on the graph to extract k-mer relationships and learn embeddings.

Main Results:

• Kmer-Node2Vec demonstrated a 29-fold increase in training speed compared to DNA2Vec on a 4GB dataset.
• The proposed method achieved comparable accuracy to DNA2Vec in DNA sequence retrieval and classification tasks.
• The graph-based approach effectively captures k-mer relationships for efficient embedding generation.

Conclusions:

• Kmer-Node2Vec provides a significantly faster and scalable solution for DNA sequence modeling.
• The method offers a viable alternative to DNA2Vec for large-scale bioinformatics analyses.
• Efficient k-mer embedding is critical for advancing DNA sequence analysis and applications.