Title: Dynamic network reconstruction from gene expression data applied to immune response during bacterial infection
1Dynamic network reconstruction from gene
expression dataapplied to immune response during
bacterial infection
Bioinformatics, vol21,no.8 pp1626-1634
2Motivation
- The immune response to bacterial infection
represents a complex network of dynamic gene and
protein interactions. The goal is to develop a
optimized reverse engineering strategy to aimed
at a reconstruction of this kind of interaction
networks.
3Transcriptional Regulation (Background)
Each gene has a promoter region where activators
and repressors can attach. While prokaryotic
regulation tends to be more simple and many times
involves just a single activator or repressor,
eukaryotic regulation tends to involve a
combinatorial interactions between several
transcription factors. This allows a more
sophisticated response to multiple conditions in
the environment which is then able to generate
spatial temporal differential expression.
Eukaryotes also make use of enhancers , distant
regions of DNA that can loop back to the promoter
.
4Statistical Problem
- In standard gene expression profiling there are
many more variables (N genes) than measurements
(M experiments), so the genetic interaction
matrix (NxN) of linear model cannot be uniquely
determined by measurement matrix (NXM)
5Solutions
- fuzzy clustering to reduce the number of
variables. This can be justified by presence of
regulatory motifs. - Methods for finding sparse interaction matrices
by combinatorial search strategy. - Singular value decomposition (SVD) based methods
that calculate a solution to the interaction
matrix by imposing additional mathematical
constraints.
6Strategy
- 1). Data pre-processing.
- 2). Optimized fuzzy clustering and cluster
validation. - 3). Selection of cluster-representative genes.
- 4). Reconstruction of probable dynamic network
models ( fitting the simulated kinetics to the
experimental expression profiles. )
7Dataset
- Stereotyped and specific gene expression
programs in human innate immune responses to
bacteria. - Boldrick JC et al.
- Proc Natl Acad Sci U S A. 2002 Jan 2299(2)972-7
- genome-www.stanford.edu/hostresponse/download.shtm
l
8Data pre-processing
- Data represents logarithm zed ratios (log-ratios)
of the expression intensities of 18 432 genes at
5 time points (t0,0.5,1,2,4h) before and after
an infection with heat-killed pathogenic E.coli. - A total of 1336 genes was selected by requiring
an up regulation or down regulation of at least a
factor of 8 - For clustering,the time profiles were scaled by
their respective absolute temporal extreme values
to focus on qualitative behavior.
9Clustering and validation by fuzzy C-means
(FCM)algorithm
uij membership matrix xi data entry cj center
of the jth cluster norm functions that
measure distance
10Selection of cluster-representative genes
- is assigned to one cluster with a high fuzzy
membership degree n (MSD) - annotated with a known immunological function
- is represented by an expression profile with no
missing values.
11Dynamic modeling
xj(t) expression of gene j1 to C at time t wi,j
gene interaction matrix bi external(infection)
stimulus response vector u(t) heaviside step
function
12Dynamic modeling using a search strategy
- Initialization
- combinatorial search
- First phase forward selection of genetic
interactions, model complexity expanded - Second phase backward elimination to simplify
obtained model - Third phase refining model by adapting the type
of dynamic dependency
13(No Transcript)
14(No Transcript)
15L1-fit, n31 mse1.52
16(No Transcript)
17Alternative structure of the dynamic system with
third order time lag elements for CD59 and STAT1
obtained from the proposed Network generation
methods configured by Rmax3 and Emax1
18Online resource
- Validation
- https//www.cs.tcd.ie/Nadia.Bolshakova/validation_
algorithms.html - Introduction for clustering algorithm
- http//www.elet.polimi.it/upload/matteucc/Clusteri
ng/tutorial_html/index.html
19Algorithm of fuzzy C-means clustering