Cross-platform validation and class prediction models. (A-B) Normalized miRNA expression data using paired microarray (A) or quantitative RT-PCR (B) are displayed for a randomly selected subset of ET (N = 10) and control (N = 10) platelets (each group included 5 males and 5 females). For both microarray and quantitative RT-PCR, boxes represent the within-group interquartile range encompassing 50% of the values, whereas the 95% confidence intervals and outliers are depicted; the horizontal bar within each box represents the group median. (C) Scatter plots demonstrate between-platform concordance for individual miRNAs by phenotype; the x- and y-axes represent the microarray and quantitative RT-PCR data, respectively, using a unified scale standardized to mean = 0 and SD = 1; 19 of 20 miRNAs (except for miR 1274a) exhibit concordant group medians, validating the expression data using independent platforms. *Statistically significant Spearman correlation coefficients (r) with P < .05. (D-E) Linear discriminant analysis plots display the posterior classification probability of each subject using a 3-biomarker subset (miR 10a, miR 148a, and miR 490 5p) based on microarray (D; N = 57) or quantitative RT-PCR (E; N = 20). Triangles represent samples containing the JAK2V617F mutation (either homozygous or heterozygous).