Improving Hidden Markov Models for classification of human immunodeficiency virus-1 subtypes through linear...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Profile Hidden Markov Models (pHMMs) are standard for sequence family modeling.
• Modeling small subfamilies, such as rare human immunodeficiency virus-1 (HIV-1) subtypes, remains challenging.
• Accurate HIV-1 subtyping is critical for epidemiology and treatment.

Purpose of the Study:

• To improve the performance of pHMMs for classifying HIV-1 sequences, particularly for subfamilies with limited data.
• To adapt existing HMM architectures using supervised learning for enhanced accuracy in small sample scenarios.

Main Methods:

• Employed a machine learning approach by integrating regularized linear discriminant learning with an existing HMM architecture.
• Replaced unsupervised estimation of emission probabilities with a supervised method, utilizing a balancing scheme for varying sample sizes.
• Applied a softmax function to transform discriminant weights into valid probabilities for classification.

Main Results:

• Demonstrated significant improvement in pHMM performance for HIV-1 sequence classification using the proposed supervised technique.
• Successfully classified partial-length HIV-1 sequences and semi-artificial recombinants.
• Showcased the effectiveness of the method in handling small subfamilies, a common issue in HIV-1 subtype analysis.

Conclusions:

• The supervised approach effectively enhances pHMMs for modeling biological sequence families, especially with small sample sizes.
• This method offers a robust solution for accurate HIV-1 subtyping, addressing a key challenge in bioinformatics.
• The findings suggest broader applicability of this machine learning-enhanced HMM approach in other sequence classification tasks.