90). The PC-containing models have much lower BIC scores and higher adjusted R2 values compared to all other models (row D in Table  1 and Additional file 3: Table S3). This means that the PCA is able to consolidate the relevant functional variation into fewer variables by replacing a handful of HB MM-102 supplier expression rates with a single PC and still retaining the same ability to predict rosetting. For example, relative to any individual expression rate, PC 1 appears to be a better predictor of whether an isolate will express severe spectrum phenotypes or mild spectrum phenotypes. Thus, the expression

rates of many HBs appear to be non-independent with respect to their relationships to phenotype. Our PCA results also imply that within the small DBLα tag there are multiple independent genetic components that are relevant to disease phenotype, since otherwise we would not expect to find more than one PC playing a significant role in any of the phenotype prediction models. This conclusion is consistent with the fact that many of the first several PCs Cilengitide explain similar levels of variation among isolates (Additional file 1: Figure S13 and S14). The principal components improve phenotype prediction, but they

are less straightforward to interpret than individual HB expression rates. Nevertheless, our results demonstrate that PC 1 clearly corresponds to the major division found by network analyses, severe and mild spectrum associated var genes. Furthermore, CH5424802 the various correlations between phenotypes and PCs, and between the expression rate of various sequence types and PCs, can be summarized in networks, which can provide additional means to interpret the PCs (Figure  5E; Additional file 1: Figure S11). In summary, we find that two PCs capture interesting phenotypic distinctions among isolates, and we find that model BICs improve considerably when PCs are used in place of individual HB expression rates. The consistency Etomidate of HB-phenotype associations in distinct populations HB analysis of a

smaller dataset from Mali that was originally analyzed by Kyriacou et al. [14], reveals that at least some of the HB-phenotype associations reported above are similarly informative in geographically distinct (and presumably genetically unrelated) populations. Twenty-four of the 29 HBs we identified in the Kenyan dataset (Figure  1) were present in the Malain dataset (data not shown). The Malian dataset contains 9 isolates from cerebral cases of malaria, and 8 isolates that serve as negative control for severe disease since they are from mild hyperparasitemic cases. Kyriacou et al. argue that mild hyperparasitemic malaria is the appropriate negative control for cerebral malaria since the two forms of disease exhibit comparable levels of parasitemia.