Sex and discriminant analysis based on 3-group microarray expression profiles. (A) Scatter plots by patient cohort were generated by applying a nonparametric Wilcoxon rank-sum test to determine sex differences in gene expression for each of the 432 genes on the microarray chip. The z-score represents a weighted difference between the observed and the expected rank sums for the Wilcoxon test; thus, a z-score of 0 represents equivalent median expression (male gene expression = female gene expression), whereas the dashed lines represent 95% confidence intervals that are female (positive)- or male (negative)-skewed. The individual genes that are differentially expressed by sex are delineated in each phenotypic group. (B) List of the 11-biomarker gene subset identified by stepwise discriminant analysis. P values for 3-cohort discrimination are displayed (cut off at P < .001 for simplicity). (C) Plot of posterior classification probability demonstrates the segregation of the 3 phenotypes (ET = 24 [●], RT = 23 [○], normal = 48 []) using the 11-biomarker gene subset via LDA with leave-one-out cross-validation (group means are depicted by the individual rhomboids with cross-hairs). Each symbol represents one patient microarray that incorporates aggregate expression data from the 11-gene subset. For each patient sample, the posterior classification probabilities that a given subject belongs to ET or RT cohorts are displayed by subject; the probability that a subject belongs to the normal cohort is equivalent to 1 − the sum of the corresponding ET and RT probabilities.