Investigating alignment-free machine learning methods for HIV-1 subtype classification

Area of Science:

• Virology
• Bioinformatics
• Machine Learning

Background:

• Human immunodeficiency virus 1 (HIV-1) classification into subtypes is vital for clinical management.
• Traditional sequence alignment methods are computationally expensive for large HIV-1 datasets.
• Existing alignment-free models struggle with classifying less common HIV-1 subtypes.

Purpose of the Study:

• To comprehensively analyze sequence vectorization methods for HIV-1 subtype classification.
• To investigate the impact of natural language-inspired embedding methods on HIV-1 subtype classification accuracy.
• To develop improved computational tools for HIV-1 subtype identification.

Main Methods:

• Employed alignment-free approaches for HIV-1 genetic sequence representation.
• Utilized k-mer based XGBoost models for classification.
• Applied Word2Vec embedding with support vector machines.

Main Results:

• Achieved a balanced accuracy of 0.84 with a k-mer based XGBoost model, demonstrating robust performance across common and uncommon HIV-1 subtypes.
• Word2Vec-based support vector machine models showed promising precision and balanced accuracy.
• Demonstrated the efficacy of natural language-inspired sequence vectorization for HIV-1 classification.

Conclusions:

• Sequence vectorization significantly impacts HIV-1 subtype classification performance.
• Natural language-inspired encoding methods offer a promising avenue for enhancing HIV-1 subtype classification.
• Improved classification can lead to better patient outcomes and targeted therapies for HIV-1.