Speaker
Description
HIV integration into the human genome establishes long‑lived proviruses that persist despite antiretroviral therapy (ART) and remain a central barrier to cure. Comparative HIV integration site (IS) analyses across individuals have revealed persistent features of proviral landscapes and differences associated with age at exposure, timing of ART initiation and duration on ART. However, IS datasets are extremely sparse, each person typically contributes only tens to thousands of unique IS among billions of potential genomic loci and statistical approaches must therefore support inference under limited person level observations, small sample sizes and high dimensional categorical annotations.
Traditional IS analyses generally either pool sites across individuals or model person specific integration rates using regression type frameworks but both approaches face limitations under data situations where most categories have no observations (high zero counts). Here we present three new statistical methodologies developed that address these challenges through Bayesian and fiducial principles, each with an equivalent method that requires no prior specification
First, we introduce a simple method to compare how often integration sites occur in different genomic categories across groups, even when sample sizes are very small. We apply this to pathways and to well-known integration enriched genes such as BACH2 and STAT5B. Second, we present a pattern finding approach that identifies the smallest set of genomic categories that best distinguish one group from another, functioning like an easy to interpret classification tool without requiring modelling assumptions. Third, we develop a flexible modeling framework that can handle complex datasets where many categories contain no detected integration sites, along with a simplified version for analyses focused only on whether a site appears at least once. We demonstrate these methods using simulations and apply them to an unpublished dataset.
| Expedited Notification | No thanks, I do not require Expedited Notification |
|---|
