Figure 2

Incorrect class predictions of ES and viremic samples relate to CD4 count and viral load. Eight control, seven ES, and six viremic samples were grouped as control+ES vs. viremic (A) and infected vs. uninfected (B). Seven class prediction methods were applied to these groups: Compound covariate, diagonal linear discriminate analysis, 1-nearest neighbor, 3-nearest neighbors, nearest centroid, support vector machines, and Bayesian compound covariate predictor. The number of incorrect class predictions for each ES and viremic sample is shown. Controls are omitted due to a high level of correct predictions for uninfected samples. CD4 count gradients for ES and viral load gradients of viremic samples are indicated with open and filled triangles, respectively (not to scale). ES (open bars) with the lowest CD4 counts were most likely to be misclassified with viremic samples in the comparison of control and ES with viremic samples (A). The class of ES with the highest CD4 counts was most likely to be incorrectly predicted when the two classes were infected and uninfected.