Parameter identifiability and uncertainty quantification in a delayed within-host HIV dynamical model

Accurate parameter inference in within-host viral infection models is frequently hindered by parameter coupling and limited observability, particularly when only viral-load data are available. In this study, we investigate structural and practical identifiability, posterior uncertainty, model adequacy, and predictive robustness in a delayed within-host HIV dynamics framework that incorporates intracellular infection latency. Using simulated longitudinal viral-load data, we combine differential-elimination analysis, profile likelihood, Bayesian posterior inference, information-theoretic model comparison, variance-based global sensitivity analysis, and forward uncertainty propagation. Under the known nonzero history functions used in the synthetic design, differential elimination establishes generic local structural identifiability of the seven model parameters and makes explicit the parameter combinations entering the viral-load observation map. Re-optimized profile likelihoods for the infection-rate parameter and the infected-cell death-rate parameter both cross the 95% likelihood-ratio threshold on either side of their minima, giving bounded intervals. The infection-rate profile is nevertheless broader and more asymmetric, indicating weaker practical constraint rather than practical non-identifiability. A conditional three-parameter, four-chain Bayesian analysis yields concentrated marginal posterior distributions for the infection-rate, virion-production, and intracellular-delay parameters, with split R-hat values no larger than 1.005 and pooled effective sample sizes between 935 and 1398. However, the infection-rate and virion-production parameters show a very strong negative posterior correlation of approximately -0.994, revealing substantial compensatory dependence that is not apparent from marginal credible intervals alone. Within the matched synthetic benchmark, the delayed formulation is favored over a reduced non-delayed alternative by both the Akaike Information Criterion and Bayesian Information Criterion, although this comparison is interpreted as a model-recovery experiment rather than external validation because the data were generated from the delayed model. Global sensitivity analysis identifies the infected-cell death rate and viral clearance rate as the strongest contributors to equilibrium viral-load variability, with the virion-production rate and target-cell production rate showing intermediate effects, whereas predictive uncertainty is greatest during the transient growth and peak phase. These findings show that trajectory fit alone is insufficient for reliable mechanistic inference and that structural identifiability, practical likelihood geometry, joint posterior dependence, and predictive robustness should be evaluated together.

Authors

Institutions

Publication Details

Journal
PLoS ONE
Published
2026-11-16
DOI
https://doi.org/10.1371/journal.pone.0362668
Primary Topic
Mathematical and Theoretical Epidemiology and Ecology Models
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Parameter identifiability and uncertainty quantification in a delayed within-host HIV dynamical model

Supanai Suksai, Suthep Suantai, Areeyuth Sama-Ae, Chunchom Salikupata
1 links
PLoS ONE
Mathematical and Theoretical Epidemiology and Ecology Models
article

Parameter identifiability and uncertainty quantification in a delayed within-host HIV dynamical model

Supanai Suksai, Suthep Suantai, Areeyuth Sama-Ae, Chunchom Salikupata
article en

Abstract

Accurate parameter inference in within-host viral infection models is frequently hindered by parameter coupling and limited observability, particularly when only viral-load data are available. In this study, we investigate structural and practical identifiability, posterior uncertainty, model adequacy, and predictive robustness in a delayed within-host HIV dynamics framework that incorporates intracellular infection latency. Using simulated longitudinal viral-load data, we combine differential-elimination analysis, profile likelihood, Bayesian posterior inference, information-theoretic model comparison, variance-based global sensitivity analysis, and forward uncertainty propagation. Under the known nonzero history functions used in the synthetic design, differential elimination establishes generic local structural identifiability of the seven model parameters and makes explicit the parameter combinations entering the viral-load observation map. Re-optimized profile likelihoods for the infection-rate parameter and the infected-cell death-rate parameter both cross the 95% likelihood-ratio threshold on either side of their minima, giving bounded intervals. The infection-rate profile is nevertheless broader and more asymmetric, indicating weaker practical constraint rather than practical non-identifiability. A conditional three-parameter, four-chain Bayesian analysis yields concentrated marginal posterior distributions for the infection-rate, virion-production, and intracellular-delay parameters, with split R-hat values no larger than 1.005 and pooled effective sample sizes between 935 and 1398. However, the infection-rate and virion-production parameters show a very strong negative posterior correlation of approximately -0.994, revealing substantial compensatory dependence that is not apparent from marginal credible intervals alone. Within the matched synthetic benchmark, the delayed formulation is favored over a reduced non-delayed alternative by both the Akaike Information Criterion and Bayesian Information Criterion, although this comparison is interpreted as a model-recovery experiment rather than external validation because the data were generated from the delayed model. Global sensitivity analysis identifies the infected-cell death rate and viral clearance rate as the strongest contributors to equilibrium viral-load variability, with the virion-production rate and target-cell production rate showing intermediate effects, whereas predictive uncertainty is greatest during the transient growth and peak phase. These findings show that trajectory fit alone is insufficient for reliable mechanistic inference and that structural identifiability, practical likelihood geometry, joint posterior dependence, and predictive robustness should be evaluated together.

PLoS ONEVol. 21(11)
Prince of Songkla University (TH), Chiang Mai University (TH), King Mongkut's University of Technology Thonburi (TH), Srinakharinwirot University (TH)
Chiang Mai University
Good health and well-being
Mathematical and Theoretical Epidemiology and Ecology Models
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon