Alignment-Free Machine Learning Serotype Classification of the Dengue Virus.

Area of Science:

• Virology
• Genomics
• Bioinformatics

Background:

• Dengue virus (DENV) serotyping is vital for epidemiological surveillance, clinical risk assessment, and vaccine evaluation due to distinct serotype characteristics.
• Current sequence alignment and phylogenetic methods are computationally intensive and struggle with fragmented or error-prone sequences common in diagnostics.

Purpose of the Study:

• To develop a fast, alignment-free method for accurate DENV serotyping.
• To create a scalable and transparent approach suitable for real-world diagnostic and surveillance data.

Main Methods:

• Utilized compact 3-mer composition features from DENV sequences.
• Employed a lineage-aware Random Forest classification framework with ambiguity masking.
• Validated using strict cluster-aware controls at 99% and 95% nucleotide identity thresholds on internal and independent external datasets.

Main Results:

• Achieved near-perfect accuracy and macro-F1 scores on internal test sets.
• Demonstrated 100% accuracy and macro-F1 performance on strictly independent external datasets, confirming robust generalization.
• Showed stable performance with sequence truncation down to 300 bp and in the presence of ambiguous nucleotides.

Conclusions:

• A lightweight, alignment-free machine learning approach using 3-mer composition rivals alignment-dependent methods for DENV serotyping.
• The framework offers high predictive accuracy, computational efficiency, and robust validation, suitable for genomic surveillance and diagnostics.