NUS School of Computing
Postgraduate Seminar by

LIU Tie Fei

LEARNING GENE NETWORK BY BAYESIAN NETWORK FRAMEWORK

Tutorial Room 9, SoC-1 #06-15 

2 May 2006, 3.00pm

Abstract:

Learning gene networks is one of the central problems in molecular biology.
In recent years, with enormous microarray data becoming available, 
learning gene network has received increasing attention, becoming one 
of the hottest topics in computational biology. Many models and learning 
methods have been proposed to solve the problem. However, the data 
problem and the complexity of gene regulatory systems make learning 
difficult. Moreover, some important biological factors which are critical
to gene regulatory systems are not considered in most published works. 
There factors include: various time delays among gene regulatory systems, 
the effects of complexes and the effect of proteins as hidden variables 
when learning a gene network from microarray data. In this thesis, 
three models and learning methods are proposed to take into account the 
important biological factors: 

1) The time delayed model is proposed to capture the various time delays 
among gene regulatory systems. A corresponding learning algorithm, 
the Time Delayed Network Learning (TDNL) algorithm, is proposed to 
learn the structure of a network. 

2) Conditional dependance is used to find the collaborations among 
regulators and the effect of a complex is considered by the learning algorithm, 
the Conditional Dependance (CD) learning algorithm. 

3) Proteins are modeled as hidden variables in the network by 
Semi-Fixed Network. The Semi-¯xed Structure Expectation Maximization (SSEM) 
algorithm, is proposed to learn the structure of a network. The effectness 
of the proposed methods are veri¯ed by experiments on both artificial and 
real-life gene expression data. The performance comparison of these methods
against some published methods prove the advantages of the proposed methods.