Abstract
With a goal of unveiling mechanisms by which vaccines can provide protection against HIV-1 acquisition, several studies have explored correlates of risk of HIV-1 acquisition in HVTN 505, which was a phase IIb trial conducted to assess the safety and efficacy of a DNA plasmid and recombinant adenovirus serotype 5-vectored HIV vaccine regimen among individuals in the United States who were vulnerable to acquiring HIV. While this trial failed to meet its predetermined efficacy criteria, both immunological and virological correlates of reduced risk of acquisition have been reported, suggesting that at least some vaccine recipients were protected from some viruses. In this work, we describe application of a novel Positive-Unlabeled machine learning-based approach to infer protection status among vaccine recipients that did not acquire HIV, resulting in improved power to detect potential correlates of immunity. Having established the analytical robustness of protection status predictions using cross-validation and permutation testing strategies, we report increased confidence in previously identified correlates of risk, such as vaccine-elicited anti-HIV-1 Env glycoprotein IgG3 antibodies and antibody-dependent phagocytosis, and the new observation of an inverse correlation between inferred vaccine-mediated protection and virus-specific IgA responses. Though its biological validity is not established, this inference approach offers a new means to use case-control datasets to identify candidate markers of effective immune responses in the context of low vaccine efficacy.