2018  •   Computer and Information Sciences

A “Systems Serology” Approach to Elucidating Mechanisms of HIV Control

Lead Presenter: Anush Devadhasan

PI: Galit Alter

Faculty Advisor: Jishnu Das

Winner


Computer and Information Sciences

Of particular research interest in immunology is infection in HIV controllers: patients exhibiting immunological suppression of AIDS, in the absence of antiretroviral therapy. Prior work has failed to identify the underlying immunological mechanisms, which would pave a path towards an HIV vaccine. One key limitation of current approaches is a disproportionate emphasis on antibody (Ab) titer/neutralization, rather than systematic analysis of the full repertoire of Ab effector functions. We employ a “systems serology” approach to address this.

We analyze a cohort of 78 human subjects comprising four distinct patient groups exhibiting varying degrees of HIV control, and one control group. We quantify the entire humoral response via functional and biophysical assays to be used as features to build a discriminative model for patient group stratification. A model utilizing an ensemble strategy comprising two random-forest classifiers (1. functional and glycan features 2. biophysical features) yielded maximum performance. A median classification accuracy across 100 independent five-fold cross-validation runs of ~ 0.7, with P=1e-4 compared to a null model, is achieved.

Out-of-bag estimates of variable importance are then computed to identify key features. We find that five functions – Ab-dependent neutrophil phagocytosis, cellular cytotoxicity and NK cell activity, and three biophysical features are most informative in HIV controller status discrimination. Furthermore, betweenness centrality and degree distributions in correlation networks of humoral features reveal that low viral load is associated with greater mechanism stability. These insights can, therefore, be leveraged in a wide range of downstream studies, including rational HIV vaccine design.

View Poster