Fig. 7: Predictive capabilities of the 9 candidate biomarkers in identifying falciparum- and coatneyi-cerebral malaria.

a Multidimensional scaling (MDS) visualization of Euclidean distances between samples based on log expression levels of the nine biomarker genes. b Performance of OPLS-DA models in discriminating between samples of different biological conditions based on the expression levels of the 9 selected biomarker genes. The performance metrics R2Y and Q2Y measure the proportion of biological label (Y) dispersion explained by the model on the training data (R2Y) and in 7-fold cross-validation (Q2Y). The biological labels were permuted 100 times to evaluate model overfitting and the metrics pR2Y and pQ2 represent the proportion of opls models trained on randomly permuted labels that have better predictive performance than the model trained on the actual labels based on R2Y and Q2Y metrics respectively. High values of R2Y and Q2Y are desirable indicating better representation of the biological conditions by the biomarker genes, whereas low values of pR2Y and pQ2Y are desirable indicating less overfitting.