An improved alignment-free model for DNA sequence similarity metric.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• DNA clustering is crucial for analyzing large-scale DNA sequences.
• Current alignment-free methods often overlook valuable sequence information.
• Improving DNA sequence similarity metrics is key to enhancing clustering quality.

Purpose of the Study:

• To introduce a novel alignment-free model for DNA clustering.
• To enhance DNA sequence similarity metrics by incorporating compound information.
• To improve the overall quality of DNA clustering.

Main Methods:

• Proposed the Category-Position-Frequency (CPF) model.
• Converted DNA sequences into three category-based sequences.
• Generated a 12-dimension feature vector using an entropy-based model considering word frequency and position.

Main Results:

• The CPF model demonstrated superior performance compared to mainstream alignment-free models (k-tuple, DMk, TSM, AMI, CV).
• Achieved better clustering results and identified optimal settings.
• Validated through experiments on multiple DNA sequence datasets.

Conclusions:

• Hybrid information models outperform frequency-only models for DNA clustering.
• A sliding window size of two is optimal for the CPF model with sequences up to 5000 characters.
• The CPF model offers efficient, stable performance and broad generalization capabilities.