Non-parametric methods to predict HIV drug susceptibility phenotype from genotype

Area of Science:

• Genetics
• Virology
• Computational Biology

Background:

• Effective medical management of human immunodeficiency virus (HIV) infection relies on understanding the link between viral genetic sequences and antiretroviral drug susceptibility.
• High-dimensional viral genotype data poses challenges for traditional statistical analysis in predicting drug response.

Purpose of the Study:

• To develop novel non-parametric statistical methods for analyzing high-dimensional HIV genotype data.
• To predict viral phenotype (drug susceptibility) from viral genetic sequences.
• To identify specific genetic patterns associated with antiretroviral drug resistance.

Main Methods:

• Development of non-parametric, non-recursive methods involving three stages: model building, identification of influential amino acid patterns, and analysis of codon combinations.
• Application of these methods to a dataset of 2747 HIV protease genome sequences and amprenavir IC50 measurements.
• Utilizing forward-stepwise modeling to predict phenotype from genotype.

Main Results:

• Identification of eight specific codons in the HIV protease region that predict resistance to the antiretroviral drug amprenavir.
• Discovery of codon pairs exhibiting concordant (occurring together) or discordant (mutually exclusive) associations.
• Demonstration of the utility of the developed methods in uncovering genotype-phenotype relationships.

Conclusions:

• The developed non-parametric methods are effective for analyzing high-dimensional HIV genetic data to predict drug resistance.
• Specific codon patterns in the HIV protease gene are key predictors of amprenavir resistance.
• These findings can inform personalized treatment strategies for HIV infection by predicting drug susceptibility based on viral genotype.