Robust supervised and unsupervised statistical learning for HIV type 1 coreceptor usage analysis.

Area of Science:

• Virology
• Computational Biology
• Immunology

Background:

• Human immunodeficiency virus type 1 (HIV-1) entry into target cells depends on coreceptor usage, impacting pathogenicity and treatment.
• Predicting HIV-1 coreceptor tropism from viral envelope sequence is challenging due to high genetic variability.

Purpose of the Study:

• To investigate associations between HIV-1 envelope genetic features and clinical markers with viral tropism.
• To develop and validate a predictive model for HIV-1 coreceptor tropism.

Main Methods:

• Utilized a dataset of 2896 HIV-1 sequence-tropism pairs from the Los Alamos HIV database.
• Applied various machine learning methods, including logistic regression, support vector machines, and decision trees, for tropism classification.
• Employed bootstrapped hierarchical clustering, univariate and multivariate analysis, and cross-validation for model development and evaluation.

Main Results:

• A high-performing logistic regression model was developed to infer HIV-1 coreceptor tropism.
• The derived model achieved an accuracy of 92.76% (SD 3.07) and an AUC of 0.93 (SD 0.04).
• The model is compact and interpretable, integrating genetic and clinical features.

Conclusions:

• Accurate prediction of HIV-1 coreceptor tropism is feasible using a machine learning approach.
• The developed logistic regression model offers a reliable tool for inferring viral tropism in patients.
• This predictive capability has significant implications for personalized HIV treatment strategies and drug development.