Statistical Analysis

To compare immunoreactivity of hypothesis-generating sera against 1 SNP variant compared to its cognate, 2-tailed t tests were performed.

For the hypothesis-confirming tests, we used a log transformation of the ELISA values for the 10 peptides. Because graphically there were no large deviations from normality of homoscedasticity (Figure 1B–K), we compare the 4 geographic regions with analysis of variance (ANOVA) and compared pairs of regions with standard 2-sided t tests. We performed linear model likelihood ratio tests to adjust for the effects of age group and sex on ELISA values. We defined “immunoreactive” patients for each of the 10 values as ELISA values exceeding the cutoffs determined as the minimum values that ensure 100% specificity in the receiver operating characteristic (ROC) curves. Two-sided Fisher exact tests (Fisher-Irwin version) were used to compare the resulting binary responses between geographic regions. We present false discovery rate [21] P values adjusted (P_adj) for the fact that we tested ELISAs on 10 peptides. The pairwise comparisons present unadjusted P values. All P values are 2-sided and P < .05 is considered significant (and the use of adjusted or unadjusted P values is clear from the context). All statistical analyses and graphs were performed using Prism 6.0 (GraphPad Software, San Diego, California) or R version 3.3.1 software.

Immunoreactivity against variant peptides significantly differs by region. A, Heatmap of semisupervised clustering of the normalized optical density (OD) values, with a color scale ranging from low (dark blue) to high (dark red). B–K, Scatterplot graphs of OD for each variant peptide, with geometric means indicated by the horizontal lines. The geographic regions are indicated on the x-axis; the OD values are indicated on the y-axis. The false discovery rate–adjusted P values from analysis of variance are listed on each graph. Abbreviations: Am, Americas; CA, Central Africa; EA, East Africa; WA, West Africa.