A k-mer-based barcode DNA classification methodology based on spectral representation and a neural gas network

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• DNA barcode classification is crucial for species identification.
• Existing methods often require sequence alignment, which can be computationally intensive.
• Short DNA sequences pose challenges for accurate taxonomic classification.

Purpose of the Study:

• To propose an alignment-free method for DNA barcode classification.
• To evaluate the method's performance, especially on short DNA sequences.
• To compare the proposed method against established supervised machine learning algorithms.

Main Methods:

• Developed an alignment-free classification method using spectral representation and a neural gas network.
• Identified distinctive words (k-mers) from spectral representations for taxonomic classification.
• Compared the method's accuracy, recall, and precision against support vector machine, random forest, and other classifiers using real barcode datasets.

Main Results:

• The k-mer-based approach achieved comparable performance to other classifiers on full-length sequences.
• Demonstrated superior robustness and accuracy on short DNA fragments (200 bp).
• Achieved 64.8% accuracy at the species level with 200-bp fragments, significantly outperforming random forest (20.9%).

Conclusions:

• The proposed alignment-free method offers a significant improvement for classifying short DNA barcode sequences.
• This approach provides a more accurate and efficient alternative for DNA classification tasks.
• The method's robustness highlights its potential for analyzing fragmented genetic data.