A Role for ProteineProtein Interaction Networks in the Identification and Characterization of Potential Biomarkers - Proteomic and Metabolomic Approaches to Biomarker Discovery - page 328

TABLE 1 Summary of the Recent Efforts Using ProteineProtein Interaction Network to Identify Biomarkers (cont'd)

Disease Data source PPI interaction database Reference

Dao et al. 19 Breast cancer response

to chemotherapy

NCBI Gene Expression
Omnibus (GEO)

HPRD

Liu et al. 20

Glioblastoma,
ovarian cancer,
hepatocellular
carcinoma and leukemia

Genomewide DNA methylation
data from the Cancer Genome
Atlas (TGCA)
and ENCODE in the UCSC
genome browser

HPRD,
IntAct,
DIP,
MINT,
BIND

Borro et al. 21 B-lymphoproliferative

diseases

T-cell separation from
blood and 2D-DIGE/MS

STRING (Search Tool for the
Retrieval of Interacting Genes/
Proteins) software version 8.1

Tseng et al. 22 Gastric cancer miRNA and mRNA data

source

HPRD

Wang et al. 23

Lung cancer Microarray from GEP

database

BioGRID,
HPRD

Chang et al. 24 Liver metastasis of

gastric cancer

cDNA microarray UniHI database

Chavez et al. 25

Resistance to cisplatin
drugs in hela cells

Proteome analysis using
SILAC

STRING 8.3

Wu et al. 26

Type 2 diabetes Microarray data from NCBI

gene expression omnibus

HPRD

Barrenas et al. 27 Disease model List of complex human

disease associated genes
identi

HPRD

ed by GWAs

Goh et al. 28 Disease model Online Mendelian Inheritance

in Man (OMIM) database

Yeast 2-hybrid
experiments

provide mechanistic insight into the disease
helping to generate new and testable hypothesis.
Several recent studies have taken the approach
of combining primary data with protein e protein
interaction network data obtained from publicly
available databases, with the aim of identifying
novel biomarkers with better predictability
( Table 1 ). There is a common theme in the meth-
odology used in these studies ( Figure 1 ). Target-
ing a particular disease, these studies acquire
samples from both nonaffected individuals
and patients or from patients with different
outcomes. The

through sequencing. At this stage, the data is
clustered based upon the prevalence of a partic-
ular variable in either of the sample populations.
Second, a network is constructed or obtained
from the existing literature. The network that is
used will re ect the goal of the experiment; for

example, if apoptosis is thought to be involved
in the phenotype being studied, a network of
apoptosis proteins is constructed. Third, the
primary data and selected network are merged.
The goal is to determine whether proteins cluster
in a manner that can predict the source of the
patient sample (i.e., normal vs. disease). The
idea is that if a group of proteins whose expres-
sion levels vary according to the status of the

first step is to acquire a primary
data set of differentially expressed proteins,
RNA concentration, or genetic variations