Systems Biology (Methods in Molecular Biology)

3.1 Data Collection,
Preprocessing,
and Analysis

Prostate cancer is a highly heterogeneous cancer and a leading cause

of cancer related death worldwide [82]. A large number of

genes and miRNAs which are associated with various signaling

cascades were found to be dysregulated in several independent

studies [83–87]. Several large-scale gene expression datasets were

deposited and publically available for research purposes. For the

construction of networks involving important factors associated

with primary and metastatic prostate cancer phenotypes, we search

published literatures along with gene expression datasets. In partic-

ular, we used GSE21032 microarray dataset available on Gene

Expression Omnibus (GEO) which contains prostate cancer

expression data in primary and metastatic states [88]. This dataset

contains 218 patient-derived samples, 98 primary tumors, 13 meta-

static tumors, and 28 normal prostate tissue samples (N¼139)

with mRNA and miRNA expression profiles. Microarray data pre-

processing was implemented by aroma.affymetrix R package

[89]. Data preprocessing consists of three stages including back-

ground correction, normalization, and summarization. The RMA

method that is the most confident approach for Exon Array data

normalization was used to gene expression data normalization. To

further analyze the normalized expression, values were transformed

to log2 scale. Differential expression analyses were conducted using

the popular limma R package [90]. In order to explore differentially

expressed genes (DEGs) and differentially expressed miRNAs

(DEMs), primary prostate tumor samples were compared to nor-

mal prostate tissue samples and metastases prostate cancer samples

were compared to primary prostate cancer samples. Absolute log

fold change greater than 1 andp-value less than 0.05 were consid-

ered as cutoff to explore differentially expressed genes and miR-

NAs. p-Values were calculated and were adjusted for multiple

testing by applying the Benjamini-Hochberg (BH) correction. In

total, we found 549 DEGs (179 upregulated and 370 downregu-

lated) in primary and 1008 DEGs (254 upregulated and 754 down-

regulated) in metastatic stages atp-value<0.05 and absolute log

fold change>1. In case of DEMs, we found 55 miRNAs upregu-

lated and 43 miRNAs downregulated in primary state and 88 miR-

NAs upregulated and 89 downregulated in metastatic state at the

same cutoff selected for the analysis of DEGs.

3.2 Construction
of Gene-Transcription
Factor-miRNA
Interaction Network
for Primary
and Metastatic
Prostate Cancer

From the datasets of identified DEGs in primary and metastatic

prostate tumors, we first predicted genes that can regulate other

DEGs as transcription factors (TF). This was predicted using

TRANSFAC database [91], which is a comprehensive and unique

database on eukaryotic TFs. Thus, we constructed a dataset of

DEGs and differentially expressed TFs in the primary and meta-

static phenotypes of prostate cancer. We then constructed a

co-expression network of DETF and corresponding target genes.

Furthermore, we connected miRNAs (DEMs) with the DETF and

Integrative Workflow for Predicting Disease Signatures 265