Probabilistic topic modeling for the analysis and classification of genomic sequences.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Biodiversity Analysis

Background:

• Genomic sequence classification is crucial for biomedical research and biodiversity assessment.
• Barcode genes are key targets for genomic identification.
• Alignment-free methods offer advantages over traditional sequence alignment techniques.

Purpose of the Study:

• To propose a novel alignment-free method for DNA sequence clustering and classification.
• To leverage k-mer representation and text mining for genomic data analysis.
• To enhance taxonomic identification using probabilistic topic modeling.

Main Methods:

• Utilized Probabilistic Topic Modeling, specifically the Latent Dirichlet Allocation (LDA) algorithm.
• Applied k-mer frequency analysis to DNA sequences, treating them as documents.
• Developed a generative model from training sequences for subsequent classification.

Main Results:

• Classified over 7000 16S DNA barcode sequences from the Ribosomal Database Project (RDP).
• Achieved performance comparable to RDP and Support Vector Machine (SVM) with complete sequences.
• Demonstrated superior performance over RDP and SVM with ultra-short DNA sequences (down to 25 bp).

Conclusions:

• The proposed alignment-free method is effective for DNA sequence classification.
• The method shows robust performance, particularly with short sequence fragments.
• Exhibits a stable performance decrease across taxonomic levels as sequence length diminishes.