1. QuantiGene Plex Represents a Promising Diagnostic Tool for Cell-of-Origin Subtyping of Diffuse Large B-Cell Lymphoma 
John S. Hall, Suzanne Usher, Richard J. Byers, Rebekah C. Higgins, Danish Memon, John A. Radford, Kim M. Linton 
The Journal of Molecular Diagnostics 
Volume 17, Issue 4, Pages (July 2015)
DOI: /j.jmoldx
Copyright © 2015 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

2. Figure 1
Signature discovery and testing by in silico analysis. All probe sets (PSs; n = 37) that mapped to the 21 discriminatory genes identified in Linton et al34 were considered and trained (naïve Bayes approach) in the Lenz diffuse large B-cell lymphoma (DLBCL) fresh-frozen (FF) data set. A: Threshold selection plots for the Linton (n = 37) PS signature at the length selected. Specificity (sp), sensitivity (se), positive predictive value (PPV), and negative predictive value (NPV) are reported. Performance metrics under cross validation were calculated at a range of dichotomization thresholds to identify the most appropriate threshold. Most scores are close to 0 or 1 and, therefore, a threshold of 0.5 was selected. B: Area under the curve (AUC) under 10 repeats of cross validation over the range of signature lengths were evaluated. Mean AUC values are reported along with corresponding 2SD intervals that depict the range of probable AUC values for future predictions, should the properties of any future data remain identical. C: Naïve Bayes probabilities [of being subtype activated B cell (ABC)] were generated for each sample using the Linton (n = 37) PS model. Panel represents receiver-operator characteristic curve for the model in FF (NuGEN) DLBCL samples35 using the Linton model. AUC in this independent validation cohort is extremely high: Naïve Bayes classifications are compared with original (published) classifications. D: Application of this model to paired formalin-fixed, paraffin-embedded (FFPE; NuGEN) DLBCL samples shows an AUC of Naïve Bayes classifications are compared with original (published) classifications.35 E: Boxplots presenting the distribution of naïve Bayes probabilities using the Linton signature in FF samples. F: Boxplots presenting the distribution of naïve Bayes probabilities using the Linton signature in FFPE samples. Tukey box-whisker parameters: horizontal bar indicates median expression; box, interquartile range; whiskers, range. Any result >1.5 interquartile range is plotted as individual outlier points. GCB, germinal center B cell; Nonsign, nonsignificant; Sign, significant.
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2015 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

3. Figure 2
Real-time quantitative PCR (qPCR) validation of formalin-fixed, paraffin-embedded (FFPE)–derived candidates. A: Confusion matrix showing the Pearson correlation coefficients of reference genes (GAPDH, ACTB, IF2B, and RS9) with each other. B: Boxplots showing the distribution of gene qPCR expression for 19 of 21 Linton signature genes, where qPCR primers could be designed within 500 bp of the 3′ end of the gene. Data show the qPCR expression of 40 diffuse, large, B-cell lymphoma FFPE samples [20 activated B-cell (ABC; red) and 20 germinal center B-cell (GCB; blue)]. Tukey box-whisker parameters: horizontal bar indicates median expression; box, interquartile range; whiskers, range. Any result >1.5 interquartile range is plotted as individual outliers (circles). C: qPCR data presented as a cluster dendrogram clustered on rows (samples) and columns (gene expression) (Spearman rank dissimilarity). Heat map colored by Z-score [low expression (blue), through white, to high expression (red)]. ABC samples are identified by dark blue and GCB samples by yellow boxes on the sample dendrogram. ∗P < 0.05.
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2015 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

4. Figure 3
Quantigene 2.0 Plex (QGP) validation of formalin-fixed, paraffin-embedded (FFPE)–derived candidates. A: Confusion matrix showing the Pearson correlation coefficients of reference (Ref) genes (GAPDH, ACTB, IF2B, and RS9) with each other. B: Boxplots showing the distribution of QGP gene expression for 21 of 21 Linton signature genes, where probe sets could be designed within 500 bp of the 3′ end of the gene. Data show the normalized QGP expression of 38 diffuse, large, B-cell lymphoma FFPE samples [18 activated B-cell (ABC; red) and 20 germinal center B-cell (GCB; blue)]; two QGP samples failed quality control (QC) and were excluded. Tukey box-whisker parameters: horizontal bar indicates median expression; box, interquartile range; whiskers, range. Any result >1.5 interquartile range is plotted as individual outliers (circles). Data are representative of three replicates, unless otherwise stated. C: QGP data presented as a cluster dendrogram clustered on rows (samples) and columns (gene expression) (Spearman rank dissimilarity). Heat map colored by Z-score [low expression (blue), through white, to high expression (red)]. ABC samples are identified by dark blue and GCB by yellow boxes on the sample dendrogram. D: Bar chart showing the Pearson correlation value comparing the expression 21 genes in 38 diffuse large B-cell lymphoma (DLBCL) samples (that passed QC on all platforms) measured on U133 Plus 2.0 array (gold standard) compared with real-time quantitative PCR (red) or QGP (blue). The t-test P value represents a statistically significant difference between real-time quantitative PCR and QGP. ∗P < 0.05 (black bar).
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2015 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions