(A) PCA of 17 cell subsets (6 NK cell subsets and 11 Th2 cell subsets) using data from 26 NA and 22 PA participants, cultured with or without peanut protein. The percentage variance explained by PC1 and PC2 are indicated. Each point represents 1 sample. (B) Euclidean distances computed based on 17 NK and Th2 cell subsets. The median values (with 95% CI) of the Euclidean distances between sample pairs are shown (left), and the median distances are displayed as a heat map (right). (C) Receiver operating characteristic (ROC) curves show the AUC, sensitivity and specificity of Lasso regression models for the training set (top) and test set (bottom). Left (data set 1) shows PBMCs from PA participants, peanut stimulated and unstimulated. Right (data set 2) shows peanut stimulated PBMCs from PA and NA participants. (D) Lasso regression model plots show the relative contribution (nonzero coefficient values) of different cell subsets in distinguishing peanut stimulated versus unstimulated PBMCs from PA participants (left), or peanut stimulated PBMCs from NA versus PA participants (right). The less contributive variables shrink to zero and are removed from the models. (E) Random forest (RF) models (using proximity scores from the training data set) were used to generate MDS plots that distinguish peanut stimulated versus unstimulated PBMCs from PA participants (top left), or peanut stimulated PBMCs from PA versus NA participants (top right). The RF models were used on the test data set to generate ROC curves showing the AUC, sensitivity and specificity. Bottom left shows peanut-stimulated versus unstimulated PBMCs from PA participants; bottom right shows peanut-stimulated PBMCs from PA versus NA participants. (F) Relative importance of different cell subsets used by the RF model to distinguish peanut-stimulated versus unstimulated PBMCs from PA participants (left), or peanut stimulated PBMCs from NA versus PA participants (right).