Вы находитесь на странице: 1из 7

CONSTRUCTION OF GENE REGULATORY NETWORK USING DYNAMIC

BEHAVIOUR
Dr.P.Ganeskumar2
Mr.S.Tamilarasu1
ganesh23508@gmail.com
tmlrss@gmail.com
1
PG Scholar, Dept. Of IT, Anna University Regional Centre, Coimbatore
2
Assistant Professor, Dept. Of IT, Anna University Regional Centre, Coimbatore
Abstract In this paper we proposed the concept of Effective construction of Gene
Regulatory Network using swarm intelligence along temporal gene expression data, Gene
Regulatory Network reveals the fundamental cellular process, and provide the basic gene
relationship among the genome. Swarm Intelligence comprises the collective behavior of
decentralized, self-organised system, we adopt two swarm intelligence based methods
namely bat algorithm (BA) and ant colony optimization (ACO). In addition to these,
recurrent neural network (RNN) is employed to achieve the dynamic behavior of gene
regulatory network. Specifically ACO is used for searching the discrete space in the
network architectures and Bat for continuous space search in the gene network. We also
use Bat algorithm for train the RNN parameter for achieve the dynamic behavior of gene
network.
Keywords-Gene Regulatory Network (GRN), Bat Algorithm (BA), Ant Colony
Optimization (ACT), Recurrent Neural Network (RNN).
1 INTRODUCTION

exhibit a strong autocorrelation between

With the recent development in

successive points. Gene can be expressed

microarray technologies made accumulation

each and every time[5], the gene expression

of large amount of biological data ,

is one where gene information is converted

obtaining information from these data is

into the structures and function of a cell by a

very difficult because there is no formal

process of producing a biological functional

method for processing these data. Where

molecule of either protein or gene product.

biological data mainly consideration of gene

Gene expression is assumed to be controlled

samples,

of various points in the sequence leading to


In this paper mainly deal with time

protein synthesis. The gene regulatory

series data for achieving the dynamic

process the cells use to turn the information

behavior of gene network, time series data

on genes into gene product. The gene

regulatory process the cells use to turn the

change

information on genes into gene product. The

environmental factor. Boolean Network can

gene are transformed into an executable

be

compounds (proteins) but undergoes various

probabilistic relationship using some set of

phases during transformation; transcription

random variable . A gene network can be

factors specify the rate of conversation of

represented in weighted matrix . Then

gene into proteins. Depending on the change

recurrent neural network is used in gene

the transcription factor can be classified into

regulatory network for processing time

activator or regulator.Gene Network show

series data . The curse of dimensionality

the relationship among the genes.

where gene can be represented by gene

Needs of Gene Regulation

cluster .The RNN training can be done using

Depending on the needs of the body some

backpropagation through time (BPTT) [11]

genes have to be transcribed whereas the

from that we can understand the structural

rest have to be switched off. Helps to adapts

and functional parameter of regulatory

to the change in environment. Helps in

networks. The curse of dimensionality

differentiation in cells. Performance of

where gene can be represented by gene

various functions in the body.

cluster.

2 BACKGROUND

parameter optimization can be done using

In classical approaches where finding the

genetic algorithm [6].

gene relationship among the genes is main

3. EXISITING SYSTEM

objectives [2, 3].with the rapid development

In existing system they used particle swarm

microarray [4, 5] produced rich amount of

optimization (PSO) algorithm for training

information of gene expression activities.

the

Different kind of analytical methods have

parameters of gene regulatory network[1].

been proposed to estimate gene regulatory


network.

of

gene

represented

In

recurrent

expression
graphically

recurrent

neural

neural

network

and
using

network

(RNN)

The particle swarm optimization (PSO) is


very slow convergence in refined search

In Boolean Network the state of gene is

stage and it have local search ability Where

represented by a Boolean variable (ON or

PSO algorithm possesses great in global

OFF) and the interaction among the gene

search where WI is large and also cause

represent

premature phenomenon.

by

Boolean

function

[6].Differential equation model describe the

4. PROPOSED SYSTEM

It consists of single layer neuron, where

The proposed system for the construction of

each neuron feeding the output signal to the

gene

inputs of all other neurons

network

from

temporal

gene

expression data comprises the following

Where expression level of gene varies

steps.

temporaly

1 The RNN formalism


2. BA for training the RNN models.
3.

ACO

for

generating

biologically

candidate architecture
Gene Network
A
1
1
1

A
2
1
1

Fig 1 Gene network representation


A1 & A2 represents Activator
R represents Repressor
T represents Target
Gene network can be represent as a directed
graph G= (V,E), where each vertex V
represents a genes and each edge represents
the regulatory influence of genes. It can be
represented as an adjancey matrix In
adjacency matrix if it is 1 mean is activator
or if it is o mean it is repressor.
Recurrent Neural network

t=0
t=T
t=T
Fig 2 Description of genetic network
through RNN model. The Recurrent Neural
Network

provides

the

self-adaptive

technique to the network to make decision


and anlaysis itself. These parameters are
very adaptive in formalization of neural.
BAT ALGORITHM
Bat algorithm is an metaheuristic algorithm,
it is mainly based on echolocation behavior
of microbat with varying pluse rate of
emission and loudness [7], It is very difficult

to train the RNN parameters ,so we propose

Where xi (t) and x^i(t) are the

different training strategies,Bat algorithm

actual and predicted expression

(BA) to determine the unknown network

leve1

parameters.

Bat algorithm is applied separately

Bat use echolocation to sense distance, they

to each independent problem.

also know the difference between food or

ACO ALGORTITHM

prey in a magical way.Bat fly with random

Ant Colony Optimization (ACO) is a

velocity vi at position xi with a fixed

Meta heuristic combinatorial optimization

frequency fmin varying wavelength and

technique that mimics the foraging behavior

loudness A0. They can easily adjusted the

of Ants. ACO maintains a colony of ants and

wavelength also. And loudness can from A0

set of permissible ranges associated with all

to a minimum cost Amin.

possible discrete values of the design

Initialization of the bat population is based

variable. Each ant is allowed to choose a

on random search.

permissible range that represents the path.

Qi(t)=Qmin+(Qmax-Qmin)U(0,1)--------------(1)

Once all ants in the colony choose its path,

Where U (0,1) is a uniform distribution.

then the possible discrete value associated

X (t)=best+Ai(t)(2U(0,1)-1)----------------(2)

with the path is taken as the candidate value


for ants. At each iteration, a new colony of

Where is the scaling factor.

ants along with the set of permissible ranges

We apply problem decomposition strategy

is formed by selecting the more fit ants and

[13] the real problem is divided into N

the permissible ranges based on a particular

(N+2) RNN parameter, here evaluating

selection strategy. Some members of the

RNN parameter minimize the error between

newly formed ants and the permissible

actual and predicted time series.

ranges undergo pheromone updation to form

The objective of parameter estimation is to

new combination of candidate solution. This

minimize the error.

process

is

combination
i=1/T

2--------------------- (3)

repeated
of

until

candidate

an

optimal

solution

is

reached. ACO starts with initialization of


parameters and ants along with permissible
range. Each ant and its permissible range
are processed to construct path.

The

constructed path is explored to form

Update pheromone matric T with

different combination of possible discrete

M^,^

values for the decision variable and the

Update pheromone evaporation

objective function is evaluated.

Pheromone evaporation

Check

whether the optimum solution is reached or

End for

not. If yes stop the process otherwise the

Return solution (M,)

best path is chosen from the evaluated


values and the pheromone updation is

5.Training SOS Data Set

carried out on the best path to form new set

In this section, the ACO/PSO framework,

of permissible ranges for the next iteration.

incorporating the eDSF model of generating


candidate architectures, is applied to a real-

Algorithm 1: ACO Algorithm


Initialize

pheromone

matrix

T=[Tij]N*N

world gene expression data set for the

Intialize global adjacency matrix

purpose of reconstructing the underlying

M= [0] N*N

gene regulatory network. The data set

Intialize

global

error

vector

captures the dynamical behavior the SOS

=[]1*N

response system of E. coli, a transcriptional

for each ACO step

network

Initialize

local

adjacency

matrix

The graph shows the difference in the actual

M=[0]N*N
Initialize

6. Result

local

error

vector

and predicted value that gives the accuracy


level of the samples.

^=[]1*N
for each artificial ant k do
Generate

candidate

architecture

with adjacency matrix Mk


Obtain error vector k and Mk
Update M^,^ with MK,K
end for
Fig 3 The prediction MSE value is 1.2*10-2

of PSO algorithm

Biologically
Kyriakos

plausible

Architectures:

kentzoglanakis

and

Matthew

poole.Computional Biology voL 9:


2. Inference of Genetic Regulatory Network
from Time series Gene Expression Data.
International Joint Conference on Neural
Network: Xu R,Hu Wunsch DC 2004
Fig 4. The prediciton MSE value is 0.7*10- 1
of Bat algorithm
From the above two MSE value the accuracy
is

more

in

the

bat

algorithm.The

performance and efficiency is comparatively


high.It take very less number of iteration to
give the best value.
We have lot of swarm technique but the
efficiency is high in the bat algorithm.
In this paper we successfully implement the
bat algorithm (BA) in the gene regulatory
network for achieving dynamic behavior
.The performance is high in that, it is very
effective in training RNN parameters and
also use ACO algorithm for effective gene
regulatory network construction.
We can improve these predictions by
implementing fuzzy base bat algorithm, from
that accuracy level can be increased and
effective gene network can be constructed.
References
1. A Swarm Intelligence Framework for
Gene

Network

gene

regulatory

for

modeling

.BMC

Bioinformatics 2007, 8(supp1 6):S9


4. DeRisi JL, Iyer VR, and Brown PO:
Exploring the metabolic and genetic control
of genetic expression on a genomic scale.
Science 1997,278,533
5. Spellman P, Sherlock G, Zhang M, Iyer V,
Anders K, Eisen M, Brown P,
Comprehensive

7. Conculsion and Future work

Reconstructing

3. Schilt T, Brazma A: Current approaches to

cycle-

regulated

identification
genes

of

of
the

cell
yeast

Saccharomyces cerevisiae by microarray


6. R Xu, D.c, Wunsch R. L. Frank,
Inference of Genetic Regulatory Networks
with Recurrent Neural Network Models
Using

Particle

Swarm

Optimization,

IEEE/ACM Trans. Computional Biology and


Bioinformatics, vol 4, no 4, pp .681-692 oct
Dec 2007
7. A Hybrid Bat algorithm: Iztok Fister,
Dusan

Fister.

IEEE/ACM

Trans.

Computional Biology and Bioinformatics,


vol 5, 78-oct-Dec 2012.

Вам также может понравиться