Analysis of gene expression patterns of subjects with consistent PTSD phenotype. Shown are analysis results of expression profiles measured from PBMC samples obtained from subjects with consistent phenotypes at the time of admission to ER (a–c), and 4 month after the trauma (d–f). (a, d) Unsupervised hierarchical clustering of expression profiles of 4512 active genes at ER only (a) or at 4 months after trauma (d). In each case, the dendrogram of samples built by the clustering procedure is shown on top, annotated by the subject number and an indication of the clinical condition of the subject and the time the sample was taken by colored boxes (dark blue—a control subject at ER; light blue—a control subject at month 4; brown—a PTSD subject at ER; yellow—a PTSD subject at month 4). The expression values are displayed in the bottom panel (columns—samples; rows—genes), where yellow indicates increased expression and blue indicates decreased expression relative to the average expression of the gene in all the arrays in the study. Clustering of ER samples (a) distinguishes almost perfectly between the classes of samples, with one misclassified sample. Clustering of samples from month 4 alone (d) distinguishes perfectly between PTSD and control samples. (b, e) Overabundance plots evaluating the statistical significance of the number of genes that are differentially expressed between PTSD and control samples. Red line—the number of differentially expressed genes that separate PTSD from control samples (y-axis) that were scored a given P-value (x-axis) in a consensus of three statistical scoring measures; dark gray line—the number of genes expected by chance with that P-value from 1000 simulations with random reshuffling of subject labels; light gray area—the range of numbers of differentially expressed genes in the 95th percentile of 1000 random simulations. The dotted vertical line indicates the P-value 0.05, the threshold for defining a gene as differentially expressed. Overabundance of differentially expressed genes is observed when using only the ER samples (b) or only samples taken at month 4 (e). (c, f) Evaluation of the supervised classification of subjects' phenotypes. Top: Expression profiles of differentially expressed genes. Labels and colored boxes indicate subjects and sample time. Bottom: Classification results from leave-one-out crossvalidation. The procedure evaluates the classifier's prediction at each sample by removing it from the training data, selecting differentially expressed genes based on the remaining samples, training a classifier, and then applying it to the removed sample. Each bar indicates the prediction obtained for the corresponding sample, when holding it out. The sign of the outcome value indicates the predicted phenotype (positive—control; negative—PTSD), and the magnitude indicates relative confidence. Red cross marks denote misclassified samples. The P-value of this classification is evaluated using 1000 simulations with random reshuffling of subjects. The classification of samples at ER succeeds in nine of 11 samples (c), and at month 4, it succeeds in eight of nine samples (f). Additional abbreviations not defined in the text: PTSD, post-traumatic stress disorder; PBMC, peripheral blood mononuclear cells; ER, emergency room. For more information on this topic, see the article by Segman et al on pp 500–513.
This is a preview of subscription content, access via your institution
