Figure 12: Immunoruler diagram showing the training risk scores for 18 subjects with low (≤4.44 µg/L) and 18 subjects with high (≥9.98 µg/L) urinary Nickel. The risk scores are obtained by projecting the quantile normalized, median-summarized intensities which correspond to glycan signature GID = 191, 264, 133, using multivariate logistic regression. The projection bias is determined under the assumption of equal cost of false positive and false negative rates. In order to facilitate the interpretation of data, the scores are sorted in ascending order for each sample and colored accordingly: low Urinary Ni in blue (left bars), the high Urinary Ni in magenta (right bars). Bars with different color shades represent quartile regions. The bar intensities correspond to the probability of belonging to the high urinary Nickel group, given the training data. This kind of visualization explicitly shows the number of false positives FP = 1, false negatives FN = 2, true positives TP = 16, and true negatives TN = 17, all obtained using the cutoff value 0.5. Consequently, specificity is Sp = 94.4% and sensitivity Sn = 88.9%. The training precision is 91.7% and the observed AUC value is 0.966.