Assessing the Uncertainty in Multiple Alignments of HIV Sequences, with Implications for Vaccine Development

• Activity Number: 3
• Type: Invited
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Paul T. Edlefsen*+
• Companies: Fred Hutchinson Cancer Research Center
• Keywords: HIV ; vaccine ; clinical trial ; sieve analysis ; profile hmm ; hidden markov model

Abstract

A major challenge in HIV vaccine development is developing assays to detect vaccine-induced evolutionary pressure on infecting HIV strains. Using so-called 'sieve analysis', Peter Gilbert and colleagues showed that the (failed) "STEP" vaccine trial did have an effect, as measured by amino-acid differences between HIV strains found in infected placebo recipients versus infected vaccine recipients. In this analysis, and in other important analyses for HIV vaccine research, the hand-generated multiple alignments of HIV sequences are accepted as true and error-free. In analyzing the more recent (and borderline statistically significantly efficacious) "Thai" trial, concern about the dependence on a single multiple alignment is particularly justified: in this and future vaccine trials, a primary goal is to elicit antibodies to the extremely-rapidly-evolving envelope protein. I will show that the standard sieve analysis methods are sensitive to the alignment in this context, and will present a model-based alternative using Profile Hidden Markov Models. This approach incorporates some of the alignment uncertainty into the analysis.