Prediction of VRC01 neutralization sensitivity by HIV-1 gp160 sequence features

Area of Science:

• Immunology
• Virology
• Computational Biology

Background:

• The broadly neutralizing antibody (bnAb) VRC01 is under investigation for HIV-1 prevention in the Antibody Mediated Prevention (AMP) trials.
• Sieve analysis is employed to determine how VRC01 prevention efficacy (PE) correlates with specific HIV-1 envelope (Env) amino acid (AA) sequence characteristics.

Purpose of the Study:

• To develop predictive models for VRC01 neutralization based on Env AA sequence features.
• To rank AA features by their predictive importance for VRC01 neutralization using classification and regression methods.
• To inform the design of a statistically powered primary sieve analysis for the AMP trials.

Main Methods:

• Modeled VRC01 neutralization against 611 HIV-1 gp160 pseudoviruses from the CATNAP database.
• Applied machine learning algorithms, including Super Learner (a nonparametric ensemble method), with cross-validation and hold-out validation.
• Utilized classification and regression approaches to predict neutralization outcomes and identify key predictive features.

Main Results:

• The Super Learner model achieved an average validated AUC of 0.868 for predicting VRC01 resistance (dichotomous outcome).
• Quantitative VRC01 neutralization (log IC50) was predicted with an average validated R2 of 0.355.
• Key features predicting neutralization sensitivity/resistance included surface-accessible residues in VRC01/CD4 binding sites, Env/gp120 length, cysteine count, and N-linked glycosylation sites.

Conclusions:

• The developed machine learning framework effectively predicts VRC01 neutralization based on HIV-1 Env sequence features.
• Identified specific Env AA features that significantly influence VRC01 efficacy, crucial for primary sieve analysis in AMP trials.
• This modeling approach may also guide VRC01's application in treating existing HIV-1 infections.