Different features identified by machine learning associated with the HIV compartmentalization in semen

Area of Science:

• Virology
• Computational Biology
• Genetics

Background:

• Genetic compartmentalization of human immunodeficiency virus (HIV) in semen is a known phenomenon.
• Understanding the genetic signatures driving this compartmentalization is crucial for controlling HIV reservoirs.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for distinguishing HIV sequences tropic to semen.
• To identify key viral genetic features associated with HIV compartmentalization in semen.

Main Methods:

• Collected 2071 partial HIV env sequences from paired blood and semen specimens of 42 individuals with HIV (Subtypes B and C).
• Utilized genetic compartmentalization tests to classify sequences into compartmentalization and no-compartmentalization groups.
• Constructed an ML metadataset using AAIndex metrics for amino acid biophysicochemical properties and applied five classification algorithms.

Main Results:

• The ML model achieved high accuracy in distinguishing the compartmentalization group for Subtype B (0.87) and Subtype C (0.74).
• The model identified six significant env features related to CD4 binding, glycosylation, and coreceptor selection differentiating blood and semen proviruses.
• These identified features are strongly associated with viral compartmentalization in semen.

Conclusions:

• A novel ML approach effectively distinguishes semen-tropic HIV based on env sequences.
• The identified genetic features provide insights into the mechanisms of HIV compartmentalization in semen.
• This work advances understanding of the HIV reservoir and informs future eradication strategies.