Skip to main content
Fig. 4 | Journal of Neuroinflammation

Fig. 4

From: Mass-spectrometric profiling of cerebrospinal fluid reveals metabolite biomarkers for CNS involvement in varicella zoster virus reactivation

Fig. 4

Selection and internal validation of biomarkers for Z. meningoencephalitis. a Hierarchical clustering analysis based on discriminatory ability of biomarker candidates. The number of features was reduced by including only the 60 analytes that were selected at least once by jackknife validation. The dendrogram was generated using the complete linkage method and AUCs (binary ROC analysis) of the selected analytes as input. The fill colors correspond to AUCs, as indicated in the legend within the panel. Analytes were ordered along the y-axis in descending order according to mean AUCs of the five possible paired comparisons involving Z. men_enc. Clade marked green: all comparisons featuring Z. men_enc. Subclade marked light blue: Z. men_enc vs. the virally infected groups. Subclade marked dark blue: Z. men_enc vs. the two non-inflamed control groups. The rectangle indicates the five best classifiers for Z. men_enc. The black circle identifies the clade containing the remaining comparisons involving the milder VZV reactivation forms. b Jackknife cross-validation. The dendrogram was constructed using the frequency with which a marker was selected (0 = never selected; 1 = always selected) among the 10 best classifiers for a given paired comparison. The rectangle identifies the four markers best validated for all paired comparisons. c–f Box/scatter plots depicting concentrations of the four biomarkers for Z. men_enc selected in Fig. 4b. Each dot represents one sample; the lower and upper hinges correspond to the first and third quartiles, respectively; the whiskers indicate 1.5 times the interquartile range to the lowest and highest value, respectively; horizontal lines indicate the median. Significance of concentration differences across all sample groups was determined with Kruskal-Wallis analysis (Benjamini-Hochberg correction for multiple-hypothesis testing, P = 0.014 as significance cutoff). c SM C16:1; P = 0.00114. d Glycine; P = 0.000568. e PC ae C34:0; P = 0.00568. f lysoPC a C26:1; P = 0.00852. Significant between-group differences are shown in the panels (Mann-Whitney U test, uncorrected P values, *< 0.05, **< 0.01, ***< 0.001)

Back to article page