Estimating HIV-1 Viremic Time from Reservoir Sequence Diversity with Uncertainty Quantification.

Abstract

Accurate estimation of HIV viremic time is valuable for understanding reservoir dynamics, and informing cure trials. Traditional approaches rely on serological assays or CD4 counts, which can lack quantitative precision. Sequence-based estimates using diversity in pre-treatment plasma RNA address this limitation, but are increasingly limited in the era of immediate ART initiation. We developed Bayesian models to predict viremic time using sequence diversity from HIV reservoir sequences in two cohorts from Rakai, Uganda and Sweden. We computed six diversity metrics for gp41, RT, and matrix p17 regions. We fitted 36 Bayesian models per region using slope-fitting and weighting strategies, and evaluated them based on predictive accuracy and model diagnostics. The best-performing models were validated on participants with known diagnosis dates, but unknown seroconversion dates. Reservoir diversity increased with viremic time across all metrics. Models based on diversity from unique RT and gp41 sequences performed well, with improved predictions when combined, particularly using simple diversity metrics such as nucleotide diversity and mean TN93 distances. In validation, these models produced estimates and credible intervals that aligned with known HIV diagnosis dates; and inclusion of weights using log-transformed sequence counts increased the precision of prediction. Models using matrix p17, complex diversity metrics, or identical sequences showed weaker performance. We present a new Bayesian approach to estimate HIV viremic time from reservoir sequences with associated uncertainty estimates. Our prediction approach works across subtypes and chronic infection, uses simple diversity metrics, and may support research on HIV reservoir dynamics and cure.