hivcopred

hivcopred predicts HIV-1 coreceptor usage from amino acid sequences of the V3 region of gp120 to distinguish CCR5-using (R5) versus CXCR4-using (X4) strains and inform antiretroviral decisions such as Maraviroc suitability.

Key Features:

• V3-based prediction: Uses amino acid sequences of the gp120 V3 region as the input determinant for coreceptor tropism prediction.
• Comparative composition analysis: Analyzes amino acid composition differences between 1,799 R5-tropic and 598 X4-tropic V3 sequences.
• Residue frequency identification: Reports residue enrichments with Asparagine and Isoleucine favored in R5 sequences and Lysine, Arginine, and Tryptophan favored in X4 sequences.
• SVM models: Implements Support Vector Machine models using amino acid composition, dipeptide composition, and split amino acid composition, achieving up to 90% accuracy.
• BLAST discrimination: Employs BLAST for additional classification, reporting prediction success rates of 93.16% for R5-tropic and 75.75% for X4-tropic sequences.
• Hybrid model: Combines approaches into a hybrid model yielding 91.66% sensitivity, 81.77% specificity, 89.19% accuracy, and a Matthews Correlation Coefficient (MCC) of 0.72.
• Independent validation: Validated on an independent dataset of 256 R5-tropic and 81 X4-tropic sequences with a maximum accuracy of 84.87% and MCC of 0.63.

Scientific Applications:

• Coreceptor tropism determination: Distinguishes R5 versus X4 HIV-1 strains from V3 sequences for tropism assignment.
• Antiretroviral guidance: Informs suitability of CCR5-targeting therapy such as Maraviroc based on predicted tropism.
• V3 structure–function studies: Supports research into gp120 V3 residue contributions to CCR5/CXCR4 usage.
• Method benchmarking: Provides comparative performance metrics for SVM, BLAST, and hybrid approaches in tropism prediction.

Methodology:

Comparative analysis of amino acid compositions between 1,799 R5-tropic and 598 X4-tropic V3 sequences identified characteristic residue frequencies; SVM models were trained using amino acid composition, dipeptide composition, and split amino acid composition (up to 90% accuracy); BLAST was used for additional discrimination (93.16% R5, 75.75% X4); a hybrid model was developed (sensitivity 91.66%, specificity 81.77%, accuracy 89.19%, MCC 0.72) and validated on an independent set of 256 R5 and 81 X4 sequences (maximum accuracy 84.87%, MCC 0.63).