Systems Biology (Methods in Molecular Biology)

profiles were analyzed by the ANOVA test for each CF patient

(Tables1 and 2 NMR).

Age and gender as well as antibiotic and probiotic assumption

were no confounding factors. Only for healthy children, a negative

correlation with PC1 (ρ¼0.426;t-test 2006 vs. 2011p¼0.03;

2007 vs. 2011p¼0.04) as well as a positive correlation with PC4

were found (ρ¼0.456;t-test 2006 vs. 2009p¼0.02; 2006 vs. 2010

p¼0.03; 2006 vs. 2011p¼0.003; 2007 vs. 2011p¼0.02).

After having verified the lack of effects of potentially confound-

ing factors on PC1 and PC4, a linear discriminant analysis (LDA)

was applied to the PC1 and PC4 components to develop a predic-

tive model for the classification of children in healthy or CF groups.

The model discriminated the two groups with a sensitivity, specific-

ity, and accuracy of 86%.

PC1

-20 -15 -10 -5 0 5 10 15

PC4

-10

-5

0

5

10

15

Fig. 1PCA score plot (PC1 vs. PC4) of the^1 H-NMR profiles of fecal samples obtained from 30 patients with
cystic fibrosis and 36 healthy children

Table 1
Spearman’s correlations between age and metabolic profiles in CF and healthy patients

Patients

Principal components

PC1 PC2 PC3 PC4 PC5

All 0.213 0.197 0.241 0.381 0.117

CF 0.204 0.027 0.440 0.200 0.318

Healthy 0.426 0.372 0.164 0.456 0.148

Significant values (p<0.05) are in bold

Metabolomics and Clinical Needs 333