Figure 5. Prediction of polyclonal sera neutralization breadth across diverse subject cohorts.

Heatmap of the relative binding of sera before (A) and after (B) STG-based enrichment for a set of 10 samples from a second Boston area (BIDMC) cohort against CD4bs mutant cores (green), the WT core, and the triple mutant STG core (both in gray). (C) Neutralization potency (ID₅₀) of the 10 sera samples across a panel of 18 tier 2 virus strains (each dot represents 1 unique virus strain) with subjects aligned as in A and B. Interquartile ranges and Tukey whiskers are shown; mean ID₅₀ values across the virus panel for each subject were compared between the STG sensitive (green) and insensitive (black) group, with significance defined by Mann-Whitney test. (D) A classifier was trained to predict the neutralization breadth of polyclonal samples from HIV-infected subjects in the Boston area (BIDMC) cohort (n = 10), a South African cohort (n = 19), or both, based on epitope-mapping input of sera or the STG-enriched IgG fraction. Dashed lines represent classifier performance when the complete mutant panel (red), CD4bs residues only (green), or the S, T, and G positions alone (blue) were used in model training. Black lines represent classifier performance when permuted neutralization class assignments were predicted using the whole panel or CD4bs and S, T, and G mutant subsets. Classification accuracy of models learned from considering S, T, and G positions only are noted at the top left of each panel in blue.