HIV-1 coreceptor usage prediction without multiple alignments: an application of string kernels.

Area of Science:

• Virology
• Bioinformatics
• Machine Learning

Background:

• Human immunodeficiency virus type 1 (HIV-1) entry into host cells depends on CD4 and chemokine coreceptors (CXCR4 or CCR5).
• HIV-1 exhibits significant sequence variability, yet conserved features dictate its functional phenotypes, including coreceptor usage.
• Predicting HIV-1 coreceptor usage from its envelope protein sequence is a classification problem addressable by machine learning.

Purpose of the Study:

• To investigate the efficacy of Support Vector Machines (SVM) with string kernels for predicting HIV-1 coreceptor usage.
• To compare the performance of different string kernels, including a newly proposed distant segments kernel.

Main Methods:

• Utilized Support Vector Machines (SVM) with various string kernels for classification.
• Employed a dataset of HIV-1 sequences from the Los Alamos National Laboratory HIV Databases.
• Evaluated classifier performance on a test set of 1425 examples.

Main Results:

• The SVM with the distant segments kernel achieved high prediction accuracies: 96.35% for CCR5, 94.80% for CXCR4, and 95.15% for both.
• The distant segments kernel demonstrated superior performance compared to other tested string kernels.
• Identified key features crucial for determining HIV-1 coreceptor usage.

Conclusions:

• The distant segments kernel is a novel and effective string kernel for HIV-1 coreceptor usage prediction.
• The SVM combined with the distant segments kernel represents the current state-of-the-art method for this classification task.
• The developed method and findings contribute to understanding HIV-1 tropism and variability.