To evaluate the usefulness of this

To evaluate the usefulness of this Filipin III novel panel of 6 biomarkers in detecting early-stage breast cancer, control subjects were analyzed against only stage I and II BC patients using a combination of univariate testing, chemometric analysis, and ROC evaluation. As shown in Table 3, these 6 differential metabolites were significant at the 0.05 level when comparing stage I and II BC patients to controls, as de-termined by GLM univariate testing. Moreover, an orthogonal PLS-DA (OPLS-DA) model constructed using these 6 metabolites showed ap-preciable differences between groups (Fig. 5A). Furthermore, as pre-sented in Fig. 5B, ROC analysis of the OPLS-DA model showed good classification performance (AUROC = 0.87, 95% CI: 0.82–0.92). As indicated by these results, the biomarker panel presented herein serves not only to distinguish BC patients from healthy controls but is also capable of discriminating stage I and II patients with localized disease from healthy control subjects with relatively high diagnostic accuracy, comparable to that of the all-stage cancer model.
3.3. Factor analysis of metabolic data
A secondary aim of this study was to relate detected metabolites to affected pathways. To this end, metabolite data were subjected to EFA [47]. This multivariate technique was performed on a reduced
Significant metabolites for comparison of BC patients at different stages and controls.
Stage I vs. controls
Stage II vs. controls
Stage I & II vs. controlsa
Stage III vs. controls
p FC
p FC
p FC
p FC
2-Hydroxybenzoic acid
Myoinositol
Proline
Palmitic acid
Hypoxanthine
Indole
Gentisic acid
5-Aminolevulinic acid
4-Pyridoxic acid
Cytidine
Nonadecanoic acid
Stearic acid
Agmatine
Indole-3-acetic acid
Pantothenic acid
2,3-Dihydroxybenzoic acid
Glycocyamine
a Early-stage BC is regarded as stages I and II.
Table 4
Differences in metabolites of patients between cancer stages, and different ER, PR, HER2 status.
Metabolites Cancer stages
ER status
PR status
HER2 status
Betaine
Palmitic acid
Asparagine
2,3-Dihydroxybenzoic acid
3-Indoxylsulfate
Indole-3-acetic acid
Hypoxanthine
Gentisic acid
Stearic acid
Taurine
5-Aminolevulinic acid
Proline
Pantothenic acid
Cytidine
correlation matrix of the 30 metabolites used for between-group com-parisons in order to determine pathways (i.e., factors) related to BC. Spectral decomposition of the experimental data matrix revealed a maximum of 4 factors (i.e., Kaiser criterion). Parallel analysis revealed only three factors accounted for more variance than random, permuted data (Supplemental Fig. S4). Subsequently, 1-, 2-, and 3-factor models were extracted and rotated in conformity with oblique promax and 
infomax criteria, totaling 6 possible factor models. Each model was comparatively examined for percentage of total variance explained, magnitude of factor loadings, number of variables loaded onto each factor, and potential for meaningful factor interpretation and sub-sequent factor assignment. The 3-factor infomax model yielded the most satisfactory solution (Table 5). The findings revealed that 3 me-tabolites loaded significantly (> 0.50) on the first factor, 3 metabolites loaded significantly on the second factor, and 3 metabolites loaded significantly on the third factor. These factors were found to be re-presentative of the arginine/proline pathway, fatty acid biosynthesis, and tryptophan metabolism, respectively, suggesting significant al-terations of these pathways in patients diagnosed with breast cancer.