Systems Biology (Methods in Molecular Biology)

Interestingly, from the results of Pearson’s simple and partial

correlation analysis among DA-based membership class probability,

clinical status (healthy or sick), and pancreatic insufficiency

(PI) factor, we inferred that the metabolic variations in cystic fibro-

sis are mainly associated with pancreatic insufficiency. In fact, the

correlation coefficient between DA-based membership class proba-

bility and clinical status (r ¼0.77) adjusted for the PI effect

dropped to 0.38 (Tables3 and 4 NMR), thus pointing to pancre-

atic insufficiency as the main driver of metabolism-based

classification.

In particular, PC4 was highly correlated with PI (r¼0.60),

while PC1 carries information partially independent of PI.

Table 2
Sources of variability in metabolomic data of CF patients and healthy children as measured by
analysis of variance (ANOVA)

Factors Patients

Principal components

PC1 PC2 PC3 PC4 PC5

Gender All 0.001 (0.976) 0.199 (0.657) 0.032 (0.858) 0.303 (0.584) 1.367 (0.246)

CF 1.077 (0.309) 3.309 (0.080) 0.090 (0.767) 1.296 (0.265) 1.75 (0.197)

Healthy 0.010 (0.923) 1.239 (0.274) 0.216 (0.645) 0.229 (0.635) 0.078 (0.781)

Antibiotic

consumption

CF 0.018 (0.894) 1.768 (0.195) 10.922 (0.003) 0.121 (0.731) 5.362 (0.029)

Probiotic consumption CF 3.116 (0.089) 0.015 (0.904) 2.909 (0.100) 0.199 (0.660) 0.590 (0.449)

The table reports theF-values for each factor and, in parenthesis, the correspondingp-values. Significant values
(p<0.05) are in bold

Table 3
Pearson’s correlation analysis among DA-based membership class probability, clinical status
(healthy or sick), and pancreatic insufficiency factor (PI)

DA-based membership

class probability

Pancreatic

insufficiency factor

Clinical status

(healthy or sick)

DA-based membership

class probability

1.0 0.762 0.768 (0.383)

Pancreatic insufficiency

factor

0.762 1.0 0.822

Clinical status (healthy

or sick)

0.768 (0.383) 0.822 1.0

In parenthesis, partial correlation coefficient between DA-based membership class probability and clinical status adjusted
for the PI effect is reported

334 Luca Casadei et al.