network. Downregulation of JUNB was associated with aggressive
metastatic prostate cancer. Based on the previous reports the JUNB
expression level can be used to assess aggressiveness of a prostate
tumor and to see its potential to progress to metastatic states
[110, 111].
3.4 Validation
of Molecular
Signatures Using
Kaplan-Meier Survival
Analysis
The Kaplan-Meier survival analysis was conducted for two pur-
poses: (1) Assessment of the methodology developed for creating
integrative co-expression networks; and (2) Validation of key
molecular signatures. It is important to see how a given biomarker
has a critical role in situ based on patient data.The Kaplan-Meier
survival analysis is a nonparametric method to estimate the proba-
bility that a patient survives beyond a given time; the idea is to
construct a series of probability tables for groups (patients with
different expression values) versus survival status at each time point
at which a failure (biochemical recurrence) occurs, assessing the
significance of difference between the survival plots for groups by
the log-rank test. In a sense, survival analysis was used to verify the
effectiveness of the co-expression network analysis. In fact, better
separation in the Kaplan-Meier curve for the identified key molec-
ular signatures has been used as a standard to judge the quality of
our methodology for the prediction. The Kaplan-Meier analysis
was conducted using biochemical recurrence (BCR) data which
demonstrates tumor relapse in patients for identified molecular
signatures. Samples were divided into two groups of high and low
expression values based on median expression values of each key
molecular signature. BCR data was used as a representative index
for prostate tumors progression. Among key molecular signatures
for primary prostate tumors the survival curves for HOXD10 and
PGR show significant difference between the high and low expres-
sion groups (p-value<0.05). We can expect long life without BCR
event in patients with high expression values for these two genes in
contrast to patients with low expression values. These results con-
firm HOXD10 and PGR as prognosis molecular signatures for
primary prostate tumor (Fig.11, left panel).
In the metastatic state key signatures (STAT3, JUN, and
JUNB) the survival curves for all of the key genes show significant
difference between high and low expression groups. Taking the
advantage of survival analysis in R we could take a look on mean
survival time and number of BCR events in the different groups and
key signatures (Fig.11, right panel). Based on this information,
among all key signatures the highest and lowest mean survival times
were 65.71 and 42.54 months in the high and low expression
groups of STAT3, respectively. Furthermore, the highest and low-
est BCR events were observed for STAT3 which were 10 and
25 BCR events for high and low expression groups of STAT3,
respectively (Fig.11).
Integrative Workflow for Predicting Disease Signatures 269