【Objective】 To propose a knowledge-fusion strategy for constructing gene networks

and to apply the proposed approach to analyze the gene networks for bipolar disorder (BPD). 【Methods】 The intersecting gene set between all genes in a protein-protein interaction (PPI) database and the genes in the whole genome single nucleotide polymorphism (SNP) dataset for BPD

provided by the Wellcome Trust Case Control Consortium (WTCCC)

were defined. Statistically significant epistatic gene pairs in the gene set were then obtained by using a haplotype-based full-model logistic regression and multiple testing corrections. Finally

a disease-causing gene network was constructed by using these epistatic gene pairs

and the hub genes whose connection degrees were significantly higher than theoretical ones were identified. 【Results】 By using the proposed knowledge-fusion strategy

the data dimension for the whole genome SNP data for BPD was reduced

from 482

248 SNPs to 98

157 SNPs. A total of 3

841 genes with significant epistasis identified by using the abovementioned statistical model were used to construct the underlying gene network for BPD

from which 115 hub-genes were found. Among 29 hub-genes with degrees more than 30

12 were found repeating the previous findings(PRKCA

EGFR

ESR1

ATXN1

FYN

CREBBP

TP53

AKT1

CSNK2A1

DLG1

PTN

and LYN)

while the remaining 17 genes might be novel susceptibility genes

judged by their functional involvements(SMAD3

SRC

GRB2

PIK3R1

ZBTB16

ABL1

APP

EP300

TGFBR1

SYK

YWHAZ

INSR

MAPK1

PRKCB

PRKCD

SMAD2

and SVIL). 【Conclusion】 This real data analysis demonstrates that the proposed gene network approach

which was guided by PPI knowledge

was a reliable systematic method

and would help us have a global view on the underlying molecular networking mechanisms for complex diseases

and would help us find the hub risk genes.