Table 1.
Cohort Subject Demographics.
Table 2.
Statistically significant differences in metabolite levels between comparator groups.
Figure 1.
Workflow diagram of biomarker candidate selection from both cohort 1 and cohort 2 data sets.
Abbreviations: HX, BCa negative but with history of BCa; Hema, BCa negative presenting with hematuria.
Figure 2.
Hierarchical clustering of cohort 1 samples (N-332) and all named biochemicals exclusive of drugs (N-422).
Subject BCa diagnosis (post urine collection) is indicated in the lower bar. Clustering was performed using complete linkage and Pearsons's correlation as the similarity metric.
Figure 3.
Cohort 1 derived candidate biomarker set heatmap for BCa vs. control groups.
Red fill cells indicate metabolites with higher mean levels in BCA urines than in non-BCa controls at a p≤0.05 significance. Green cells indicate lower levels in BCa relative to control urines at a p≤0.05 significance. Statistical q-values and profiling results for all other named compounds measured in cohort 1 samples are presented inn S1 Table.
Figure 4.
Hierarchical clustering of cohort 1 samples (N = 332) and the set of 25 candidate biomarkers.
Subject BCa diagnosis (post urine collection) is indicated in the lower bar. Clustering was performed using complete linkage and Pearsons's correlation as the similarity metric.
Figure 5.
Random Forest analysis of cohort 2 sample data using 25 metabolites selected from cohort.
Metabolites are rank-ordered by their mean decrease accuracy score. A higher mean decrease accuracy value indicates a greater predictive value. The 6 boxed data points represent top performing metabolites summarized in Fig. 6.
Figure 6.
Comparison of statistically significant metabolites from cohorts 1 and 2.
Comparisons are for all BCA positive urines versus combine BCA negative controls. Dark red and dark green cells represent fold differences with a p≤0.05. Light green cell with blue text represents p≤0.1. BLQ: below limit of quantitation; NA: not applicable.
Figure 7.
Receiver Operating Characteristic curves for a 6-biomarker algorithm.
An algorithm, utilizing the candidate biomarkers palmitoyl sphingomyelin, lactate, gluconate, adenosine, 2-methylbutyrylglycine and guanidinoacetate was trained using the cohort-1 data set and then tested on the cohort-2 data set. ROC curves with AUCs are displayed for the training set (A.) and the test set (B.).