Computational Methods in Biomolecular Structures and Interaction Networks
(9 Jul - 3 Aug 2007)
Jointly organized with Genome Institute of Singapore
~ Abstracts ~
Computational and experimental
Analysis of signaling pathway crosstalk in developing
tissues
Jessica Lembong, Princeton University, USA
The main goal of the talk is to illustrate how multiple
experimental (genomic, genetic, imaging) and computational
(mechanistic and data-driven modeling) are combined to
provide insights into the dynamics of living tissues. Given
the fact that the same pathways which coordinate tissue
development are deregulated in multiple diseases, this work
has broad implications for tissue dynamics in all animals.
Pattern formation in development relies on combinatorial
interactions of signaling pathways. Multiple cellular and
biochemical mechanisms of signaling crosstalk have been
characterized, but the spatial and temporal aspects of
pathway integration remain poorly understood. We show that
in Drosophila egg development (oogenesis), the dynamics of
the evolutionarily conserved epidermal growth factor
receptor (EGFR) and bone morphogenetic protein (BMP)
pathways are coordinated by a spatially distributed
feedforward circuit. The two pathways act independently
during early eggshell patterning, but become fully
integrated when EGFR assumes complete control over BMP
signaling via coordinated regulation of multiple BMP pathway
components. The dynamic consequences of this mode of pathway
integration are manifested at the genome-wide level, in the
coordinated response of the transcriptional targets of EGFR
and BMP signals. The signaling and transcriptional patterns
induced by these interactions are conserved across species
(separated by tens of million years), suggesting that
feedforward control is a general strategy for the
coordination of convergent inductive signals.
While experimental studies led us to the discovery of this
pathway coordination mechanism, computational models are
important to further understand the extent and the
importance of this regulation. In modeling signaling
pathways, we focus on three main processes: the
extracellular transport of ligand, the cytoplasmic signal
transduction, and the nuclear-cytoplasmic shuttling of
signal transduction molecules. By combining these three
modules, we have developed a model which can make
experimentally testable predictions of the spatiotemporal
dynamics of EGFR and BMP signaling in oogenesis. We will
present the results of the computational analysis of this
model and the first steps towards the experimental
validation of its predictions.
This work is done jointly with Nir Yakoby, Chris Bristow,
Trudi Schupbach, and Stanislav Shvartsman.
Dynamic transcriptional
control of gene expressions using sequential logic model (Cellogica)
Masa Tsuchiya, Keio University, Japan
Cellular signaling involves a sequence of events from
ligand binding to membrane receptors through transcription
factors activation and the induction of mRNA expression. The
transcriptional-regulatory system plays a pivotal role in
the control of gene expression. A novel computational
approach to the study of gene regulation circuits is
presented and a simulation software (Cellogica) is
developed.
Based on the concept of finite state machine, which provides
a discrete view of gene regulation, a novel sequential logic
model (SLM) is developed to decipher control mechanisms of
dynamic transcriptional regulation of gene expressions. The
SLM technique is also used to systematically analyze the
dynamic function of transcriptional inputs, the dependency
and cooperativity, such as synergy effect, among the binding
sites with respect to when, how much and how fast the gene
of interest is expressed.
SLM is verified by a set of well studied expression data on
endo16 of Strongylocentrotus purpuratus (sea urchin) during
the embryonic midgut development. A dynamic regulatory
mechanism for endo16 expression controlled by three binding
sites, UI, R and Otx is identified and demonstrated to be
consistent with experimental findings. During transition
from specification to differentiation in the wild type
endo16 expression profile, SLM reveals the three binary
activities are not sufficient to explain the transcriptional
regulation of endo16 expression and additional activities of
binding sites are required. Further analyses suggest a novel
repression effect of R during specification to
differentiation stage that is independent of UI activation.
The sequential logic formalism allows for a simplification
of regulation network dynamics going from a continuous to a
discrete representation of gene activation in time. In
effect our SLM is non-parametric and model-independent, yet
providing rich biological insight. The demonstration of the
efficacy of this approach in endo16 is a promising step for
further application of the proposed method.
Control of fibrosis by the
Triple Helix-Forming Oligodeoxyribonucleotides targeting the
promoter of Type I Collagen gene
Ramareddy Guntaka, University of Tennessee, USA
In response to injury, an evolutionarily conserved wound
healing process occurs. This response, if gone awry, can
result in a pathologic process commonly referred to as
‘fibrosis’. Fibrosis is due to abnormal accumulation of
‘extracellular matrix (ECM) proteins’, and is responsible
for causing structural alterations and loss of function of
the involved organ in the body. The main component of the
ECM is Type I Collagen. In our laboratory, we have developed
triple helix-forming oligodeoxyribonucleotides (TFOs), which
form triplex structures with the C1 region (-170 to -141
from transcription start site) of the α1(I) collagen gene
promoter. This region contains a stretch of about 30
pyrimidines on the upper (coding) strand and the
complementary purine stretch on the lower (non-coding)
strand.
Kinetic studies indicated that the antiparallel homopurine
strand (GGGAAGGAAAGGGAGGAGGGGGGAG) forms the most stable and
efficient triplexes, with a Kd of about 10-8 compared to the
parallel homopyrimidine or homopurine strands. Even the
overlapping 18-mers (AAAGGGAGGAGGGGGGAG and
GGAAGGAAAGGGAGGAGG) form triplexes very efficiently. Both
the 25-mer and the 18-mers containing the contiguous stretch
of 6 Gs readily form G-quartets in the presence of Na+
ions. Further, we showed that triplex formation results in a
significant inhibition of transcription of the α1(I) gene
both in cell-free nuclear extracts and in cells transfected
with a1(I) collagen promoter-containing plasmid DNAs.
However, mutant TFOs which failed to form triplexes also
failed to inhibit transcription, indicating that triplex
formation is essential for inhibition of transcription.
Recently, using psoralen-conjugated TFOs to cross-link the
TFO to the target promoter, we could demonstrate formation
of triplexes with the native collagen gene promoter in
stellate cells.
In order to determine whether the TFO has any inhibitory
effect on fibrosis in vivo, we used a rat model for liver
fibrosis. It has been shown that treatment of rats with
dimethyl nitrosamine (DMN) induces hepatic fibrosis, which
is mainly mediated by production of abnormal amounts of Type
I collagen by activated stellate cells. Biodistribution
studies indicated that when the TFO is administered by
intraperitoneal route, it is predominantly taken up by the
liver and kidney and in small amounts by other organs
including heart and lung. Further we showed that the uptake
by stellate cells is relatively more compared to the
hepatocytes and endothelial cells. This uptake can be
further enhanced by conjugating the TFO to
mannose-6-phosphate. In addition, we could also demonstrate
accumulation of significant amounts of the TFO in the nuclei
of stellate cells. Parallel experiments with fibrotic rats
also indicated that the TFO reached the target stellate
cells.
Administration of these TFOs into rats at 10 mg/kg body
weight had no deleterious effect on the growth of the
animals or liver weights. At 4mg/kg, the 25-mer
significantly reduced fibrosis as analyzed by the
morphometric analysis of liver tissue sections stained for
collagen with Masson trichrome stain. In the same group of
animals, the TFO significantly improved the liver function
as assessed by ALT levels. Both 18-mer TFOs also
significantly reduced fibrosis. However, for the
G-quartet-forming 18-mer, it is absolutely necessary to
block G-quartet formation for it to be effective in reducing
fibrosis.
Gene expression: molecular
mechanisms of gene transcription
Piergiorgio Percipalle, Center of Molecular Biology, Sweden
During gene expression the information contained in a
linear DNA sequence of nucleotides termed gene is decoded
into a protein, a linear sequence of amino acids. This
process is highly regulated and consists of several steps.
Initially a DNA molecule is used as template to produce an
intermediate molecule termed messenger-RNA (mRNA) which is
then further translated to yield a protein as the final
outcome. This process is universal but there are differences
in prokaryotic and eukaryotic cells mainly due to the
different extent of cellular compartmentalisation in each
cell type. In any cases, proteins are essential for cell
survival since they are involved in all cellular processes
and therefore the process of gene expression is universal.
In this lecture we are going to discuss the mechanisms
underlying the first step of gene expression termed gene
transcription. In particular we are going to emphasize the
importance of RNA polymerase, the key enzyme in the process
and how it functions to produce an mRNA molecule. In
addition we are going to discuss the role of chromatin in
eukaryotic organisms and how modifications of chromatin
structure are required for productive gene transcription.
Finally we will put the process of gene transcription in the
context of nuclear architecture and compare it with
transcription of ribosomal genes.
Gene expression: the
mechanisms of mRNA translation and protein synthesis
Piergiorgio Percipalle, Center of Molecular Biology, Sweden
In the first part of gene expression, a gene is
transcribed into an mRNA molecule. This intermediate nucleic
acid is then used as template to decode further the
information and yield a protein. The process during which an
mRNA molecule is decoded is known as mRNA translation or
protein synthesis and it is considered as the second phase
of gene expression.
mRNA translation is a very complex event that is highly
conserved in all organisms. It is carried out by very
complex molecular machineries called ribosomes and requires
a large number of regulatory factors that guarantee the
correct functioning of the process. A faulty mRNA
translation often correlates with disease. Therefore it is
very important to understand how these mechanisms operate.
In this lecture we are going to dissect how an mRNA molecule
originated during the first phase of gene expression – gene
transcription – is decoded and we are going to go in details
on how ribosomes mediate this process.
Projection of gene-protein
networks to functional space of mammalian proteome via
alternative splicing
Alexander Kanapin, European Bioinformatics Institute, UK
Informational content of genome coding sequences unfolds via functions of proteins. Alternative splicing is one of the ways of the genome manifestation into its proteome. We consider the problem of projection of genetic information into the functional space of the proteome. The later is defined as a set of all functions committed by proteins. First, we automatically created a set of functional labels of proteins attributed by conservative protein domains. A new type of relational networks between genomic DNA sequences and functional labels has been proposed. The networks has been used to analyze the acquisition of a new function in the proteome and evolutionary plasticity of the genome.
We use the InterPro database as integrative resource combining data from different protein domain databases. About 70% of InterPro entries have UniProt Keywords (KW) assigned to them. This makes it possible to assign KW ID to a protein sequence via its conservative domains. Then we consider the combinations of KW IDs as the functional labels (FL) which characterize the biological functions of the given protein. The unique set of functional labels of a proteome is considered as the functional space or functional complexity of the proteome. Then we analyzed how protein isoforms (PI), produced by alternative splicing, differ in their functional annotation. By mapping of a set of transcriptional units (TU) for human and mouse transcriptomes, produced by the FANTOM consortium, on the set of the functional labels, we construct and characterize TU/FL interconnection networks.
We created a catalog of common and unique functional
labels to both mammalian species tested based on functional
labels analysis of the transcriptome. The new type of
functional networks and statistics of FL-to-protein links
and PI-to-TUs links for human and mouse proteomes is
derived. Using statistical analysis of these networks, we
propose an evolutionary mechanism of protein function
acquisition. The process includes stages with different
evolutionary constraints. The network analysis allows usto
analyze genome-transcriptome-proteome evolutionary
plasticity. We also compare diversity of biological
functions of proteomes in different species. The functional
networks reveal a group of genes and corresponding functions
which could be attributed to an early conservative part of
the cellular machinery.
Sequence-specific interaction
of PNA with duplex DNA
Maxim Frank-Kamenetskii, Boston University, USA
Peptide Nucleic Acid (PNA) is an artificial analog of
nucleic acids carrying DNA bases and a peptide backbone. It
has pretty unusual modes of binding to DNA and RNA. Most
interestingly, PNA has a unique ability of sequence-specific
targeting double-stranded DNA (dsDNA) by invading the DNA
double helix. There are two modes of duplex invasion
complexes of PNA with DNA. Homopyrimidine PNAs, known also
as DNA openers, invade duplex DNA via triplex formation
leaving one of two DNA strands displaced and thus capable of
interactions with single-stranded oligonucleotides and PNA
oligomers via Watson-Crick pairing. Pseudocomplementary PNAs
(pcPNAs), which carry chemically modified bases, exhibit a
double-duplex mode of binding to duplex DNA interacting with
both complementary strands. The mechanism, structure and
applications of these two types of sequence-specific
PNA-dsDNA complexes are covered. In particular, the use of
PNA openers for highly specific fluorescence detection of
short signature sequences in bacterial genomes and the use
of pcPNAs for sequence-specific bending of duplex DNA are
discussed in detail.
Hybrid hamiltonian replica
exchange based on poisson-boltzmann model
Yuguang Mu, Nanyang Technological University, Singapore
A modified sampling scheme which utilizes the framework
of replica-exchange simulation with explicit solvent
molecules and replaces the system exchange probability for
jumping between different temperature replicas with the
kernel exchange probability is brought out termed as REMDhPB.
The kernel mainly includes the peptide/protein and/or
protein complexes only. The energy of the kernel is the sum
of the vacuum energy calculated by force field and the polar
solvation energy obtained from Poisson-Boltzmann model plus
the non-polar solvation energy estimated from solvent
accessible surface. Canonical distributions of the
conformations of three distinct penta-peptides were obtained
in comparison with those from standard REMD simulations.
Moreover this method has been applied to ab inito fold a
decapeptide peptide to its native beta hairpin structure
from extended conformations.
Increasing confidence of
protein-protein interactomes
Limsoon Wong, National University of Singapore
High-throughput experimental methods, such as
yeast-two-hybrid and phage display, have fairly high levels
of false positives (and false negatives). Thus the list of
protein-protein interactions detected by such experiments
would need additional wet laboratory validation. It would be
useful if the list could be prioritized in some way.
Advances in computational techniques for assessing the
reliability of protein-protein interactions detected by such
high-throughput methods are reviewed in this talk, with a
focus on techniques that rely only on topological
information of the protein interaction network derived from
such high-throughput experiments. In particular, we discuss
indices that are abstract mathematical characterizations of
networks of reliable protein-protein interactions and
indices that are based on explicit motifs associated with
true-positive protein interactions.
Large-scale inference of
condition-specific regulation using gene expression data and
the predicted transcription factor occupancy of promoters
Neil Clarke, Genome Institute of Singapore
Gene expression experiments have been performed under
many different conditions. In contrast, large-scale ChIP-chip
experiments (i.e., those involving many transcription
factors) have been performed under just a few. It is likely,
therefore, that only a fraction of condition-specific
functional binding sites have been identified. Computational
methods are required to further correlate factors,
conditions, and target genes in order to infer more
comprehensive regulatory networks, and to generate
hypotheses that can be tested by directed ChIP experiments.
We have previously developed a method for predicting the
probability of transcription factor binding to a
promoter.[1] The method models cooperative and competitive
binding in a physically meaningful manner, and appropriately
uses protein concentration as a parameter. Genome-wide
nucleosome location data has also been incorporated into the
model to improve the prediction of bound sites.[2] We are
now using this method to systematically compare predicted
binding profiles for over a hundred transcription factors to
the changes in gene expression in hundreds of microarray
experiments, and have identified many conditions under which
predicted binding is significantly correlated with gene
regulation. A joint probability analysis, using gene
expression changes and predicted binding probabilities,
further identifies the genes that are most likely to be
direct targets of the transcription factor under that
condition. This analysis recapitulates interactions inferred
from expression and ChIP-chip analyses, and makes novel
predictions that can be tested by ChIP experiments under
previously unexplored conditions.
1.Granek JA, Clarke ND: Explicit equilibrium modeling of
transcription-factor binding and gene regulation. Genome
Biol 2005, 6:R87.
2.Liu X, Lee CK, Granek JA, Clarke ND, Lieb JD: Whole-genome
comparison of Leu3 binding in vitro and in vivo reveals the
importance of nucleosome occupancy in target site selection.
Genome Res 2006, 16:1517-1528.
An overview of weighted gene
co-expression network analysis
Steve Horvath, University of California, USA
Weighted gene co-expression network analysis (WGCNA)
facilitates a systems biologic view of gene expression data.
The network framework makes it straightforward to integrate
gene expression data with other types of data, e.g. clinical
traits and genetic marker data. This talk covers several
theoretical topics including network construction, module
definition, network based gene screening, and differential
network analysis. The methods are illustrated using several
applications including i) screening for cancer genes, ii)
comparing human and chimp brains, and iii) complex disease
gene mapping. Related articles and material can be found at
the following webpage
http://www.genetics.ucla.edu/labs/horvath/CoexpressionNetwork/
Functions, networks, and
phenotypes by integrative genomics analysis
Xianghong Jasmine Zhou, University of Southern
California Los Angeles, USA
The rapid accumulation of genomics data provides
unprecedented opportunities to systematically infer gene
functions, regulatory networks, and phenotype associations.
In this talk, we develop several graph-based data mining
algorithms to integrate diverse genomics data, especially
the vast amount of microarray data in the public
repositories. A series of microarray data sets are modeled
as a series of co-expression networks, in which we search
for frequently occurring network patterns. Our integrative
approach for functional annotation provides three major
advantages over the commonly used microarray analysis
methods: (1) enhance signal to noise separation (2) identify
functionally related genes without co-expression, and (3)
provides a way to predict gene functions in a
context-specific way. Furthermore, we show that frequently
occurring co-expression clusters are more likely to
represent transcriptional modules than those clusters
derived from a single microarray dataset. In addition, we
propose the concept of "second-order correlation" which
enables us to trace the upstream events of transcription
cascades. Finally, we develop methods to systematically
identify phenotype specific network patterns and regulatory
modules.
Modeling of genetic regulatory
networks: a data-driven process
Wei Zhang, M. D. Anderson Cancer Center, USA
High throughput genomic and proteomic studies of
clinical samples have generated a large amount of data but
very little information and much less wisdom. We understand
that transcripts and proteins are interlinked but it is a
major challenge to develop appropriate mathematical models
that reveal the logical and physical relationships among the
components of the biological systems. We submit that a key
modeling criterion is that the model has to be data-driven:
it has to be able to take in biological data and produce
experimentally testable diagrams or networks. Only when this
correlation is demonstrated again and again can we reach a
conclusion that a biologically appropriate mathematical
model is born.
Our group developed a mathematical model termed
Probabilistic Boolean Network (PBN) considering the
uncertainties and probabilistic nature of biological
systems. We applied this PBN model to a set of microarray
data generated from 25 glioma tissues that were from
different stages of the cancer developments. Then we
generated two subnetworks focusing on two important genes
for glioma development and progression: vascular endothelial
growth factor (VEGF) and insulin-like growth factor binding
protein 2 (IGFBP2). VEGF is required for angiogenesis, which
is critical for providing nutrients for tumor growth. The
VEGF subnetwork revealed a number of relationships that are
supported by literature reports. IGFBP2 is overexpressed in
80% of the most advanced glioma, glioblastoma multiforme (GBM)
and contributes to glioma cell migration and invasion.
Mathematical modeling with glioma gene expression profiling
data suggested that IGFBP2 was linked to the integrin
pathway. This notion was subsequently validated by
demonstration that IGFBP2 interacts with integrin through an
RGD domain. We hypothesized that IGFBP2 is a key regulator
of glioma progression. We tested our hypothesis using a
glial-specific somatic gene transfer mouse model called the
RCAS-tva model. In this system, avian virus receptor is only
expressed in glial cells via a neuroglial-specific nestin
promoter. Genes of interest are cloned into an avian RCAS
vector and viral particles are expanded in DF1 avian
fibroblasts. When injected into the mouse brain, the viral
particles only infect glial cells and genes of interest are
only expressed in the glial cells. Our results showed that
chronic platelet-derived growth factor (PDGF) signaling
leads exclusively to the formation of oligodendrogliomas
(O). When PDGF is delivered in combination with IGFBP2,
anaplastic oligodendrogliomas (AO) form. These higher-grade
tumors are characterized by vascular proliferation,
increased cellular density, and poor survival. Also using
this model, combined activated K-Ras and Akt signaling leads
to the formation of astrocytomas. However, up-regulation of
Ras alone or Akt alone does not result in tumor formation.
Forced IGFBP2 expression in combination with activated K-Ras
leads to the formation of astrocytomas. These tumors are
histologically similar to gliomas formed by K-Ras/Akt
stimulation. Interestingly, we do not see tumor formation
when Akt and IGFBP2 are delivered simultaneously. Therefore,
IGFBP2 and Akt likely lie in the same pathway or in
converging pathways. These data show that IGFBP2 actively
contributes to tumor initiation and progression in two
lineages of gliomas. Through these functional genomic and
mathematical modeling studies, we believe we have gained
important insight into a key gene, IGFBP2, which serves as a
potential therapeutic target.
Reference:
1. Fuller, G.N., Rhee, C.H., Hess, K., Caskey, L., Wang,
R.-P, Bruner, J.,Yung, A., and Zhang, W. Reactivation of
insulin-like grwoth factor binding protein II expression
during glioblastoma transformation revealed by parallel gene
expression profiling. Cancer Res. 59:4228-4232, 1999.
2. Shmulevich I, Dougherty ER, Kim S, and Zhang W.
Probabilistic Boolean network: a rule-based uncertainty
model for gene regulatory networks. Bioinformatics
18:261-274, 2002.
3. Song SW, Fuller GN, Khan A, Kong S, Shen W, Taylor E,
Ramdas L, Lang F, and Zhang W. IIp45, an insulin-like growth
factor binding protein 2 (IGFBP-2) binding protein,
antagonizes IGFBP-2 stimulation of glioma cell invasion.
Proc Natl Acad Sci USA 100:13970-75, 2003.
4. Hashimoto RF, Kim S, Shmulevich I, Zhang W, Bittner ML,
and Dougherty ER. A directed-graph algorithm to grow genetic
regulatory subnetworks from seed genes based on strength of
connection. Bioinformatics 20:1241-7.
5. Dunlap SM, Celestino J, Wang H, Jiang R, Holland E,
Fuller GN, and Zhang W. IGFBP2 promotes gliomagenesis and
progression. Proc Natl Acad. Sci. USA (in press)
Local signaling networks
defined by quantitative morphological signatures
Chris Bakal, Harvard University, USA
Genetically identical cells can adopt a diverse spectrum
of complex shapes in order to accomplish a variety of
functions. Classical experimental approaches have identified
hundreds of unique proteins that play roles in the dynamic
remodeling of cell shape in response to upstream signals,
but there is little understanding of how these proteins are
physically organized into networks in subcellular space, and
how information flows through this sophisticated molecular
circuitry in real-time. In order to model the signaling
networks that regulate cell shape, we have developed a novel
analytical technology termed “Quantitative Morphological
Profiling” that uses 150-600 different features to describe
the morphology of single cells.
Here we describe that quantitative morphological profiling
of single cells combined with RNAi-based genetic screening
technology results in the identification of local signaling
networks with spatially, temporally, and functionally
defined characteristics that act in a hierarchical manner to
regulate cell shape and migration. These methods can be used
not only in the context of genetic screens, but also in
large-scale screens of small-molecule libraries, or screens
involving overexpression of cDNAs. Akin to gene expression
data, we can now employ morphological data for computational
approaches that aim to model the dynamic nature of signaling
networks, while the RNAi component pushes us closer to
causal mechanistic linkages.
Micro/nano crystal network: from
understanding to design of bio-functional materials
Xiang Yang Liu, National University of Singapore
High preferment functional materials consisting of
interconnecting network become increasingly important in
both sciences and technologies. It has been shown that in
many cases, crystal networks of the hybrid structure give
rise to much more superior properties than singles crystals
themselves. For instance, the special crystal network
structure of amino acids in spider silk leads to the tensile
strength several magnitude higher than single chains of
amino acids. Due to these facts, our interests are shifted
from the control of single crystals, such as size and shape
of single crystals to the engineering the crystal network.
In this contribution, new understandings on the formation
kinetics of crystal network, and the between the network
structure and the properties of the systems will be
presented. This represents a new direction in the field of
crystal growth and crystal engineering. It can also be
visualized that in the 21st century, the engineering of
crystal network will become one of the most active
directions in materials sciences. In this talk, I will
introduce the latest development in the kinetics of fiber
network formation, the correlation between the structures of
biological functional materials and the in use properties,
and the application to nanoengineering. This includes the
engineering of nano phase and ultra-functional
bio-materials.
Probabilistic models of sampling
and emergence of biological networks
Vladimir Kuznetsov, Genome Institute of Singapore
We will show that statistics of observed events in
evolving finite biological systems cannot be formally fitted
and mechanistically explained in the terms of so-called
"scale-free" network approach. However, the families of
skewed size-dependent probability distribution functions
could be used. In particular, we demonstrate that statistics
of the number of domain-to-protein links in the proteomes of
hundreds species representing all of three super-kingdoms of
life (archea, bacteria, eukaryotes) fit well to the Markov
birth-death random process models the steady-state solution
of which is approached by size-dependent Generalized Pareto
function. A parameterization of this model allows us to
associate the complexities of prokaryotic and eukaryotic
organisms with two distinct network statistics,
respectively. We also discuss other types of stochastic
evolution models with size-dependent attributes and new
applications of such skewed probabilistic models to de-noise
the experiment data in large-scale genomic experiments. We
present the methods to identify the underlying probability
functions of gene expression levels and of avidity of
transcription binding sites in the eukaryotic genomes by
available high-noisy and incomplete data.
Computational identification of
gene sets controlled by transcription factors on genome
scale
Vladimir Kuznetsov, Genome Institute of Singapore
Advances in high-throughput technologies, such as ChIP-chip
and ChIP-PET (Chromatin Immuno-Precipitation Paired-End
diTag), and the availability of human and mouse genome
sequences now allow us to identify transcription factor
binding sites (TFBS) and analyze mechanisms of gene
regulation on the level of the entire genome. Here, we have
developed a computational approach, which uses ChIP-PET data
and statistical modeling to assess experimental noise and
identify reliable TFBS for c-Myc, STAT1 and p53
transcription factors in the human genome. We present a
mixture probabilistic model and the Monte Carlo simulation
model of ChIP-PET data to define the background noise
of the sequence clustering and to identify the probability
function of specific DNA-protein binding in the eukaryotic
genome. We will demonstrate good agreement of the
curve-fitting and simulation methods which in combination
with motif search procedure not only distinguishes bona fide
TFBSs from non-specific binding sites with a high
specificity, but also provides computational basis for
further optimization of experimental parameters of the ChIP-PET
method. We will also present novel methods of estimation of
saturation of the specific binding sites, sensitivity and
reproducibility of essentially incomplete ChIP-PET data
sets. Computational integration of ChIP-PET method with
microarray expression data will be also carrying out and
using to prediction direct target genes and transcriptional
network modules.
Computer Simulation of
Biological Pathways and Network Crosstalk based on Mass
Action Laws
Yu Zong Chen, National University of Singapore
Biological systems are complex systems involving dynamic
biomolecular interactions in the context of specific
pathways and crosstalk among different pathways.
Computational simulation of these processed enables a deeper
understanding of the functional outcomes and fundamental
mechanism of the actions of biological networks. It also
enables the identification of therapeutic targets,
simulation of disease processes, therapeutic and other
effects of drug actions. This tutorial gives introductory
overview about the simulation of biological pathways and
network crosstalk based on the application of mass action
laws. The simulation model for the EGFR-Ras/MAPK pathway and
the simulation model of RhoA's crosstalk to EGFR-mediated
Ras/MAPK activation via MEKK1 are used as illustrative
examples.
Building and using protein
interaction networks: industry perspective
Andrej Bugrim, Genego, Inc., USA
In this presentation we describe our efforts in
collecting protein interaction database, developing
functional ontologies and building mechanistic
reconstructions of major biological pathways. We describe
our annotation process by which we collect information about
three major levels of cell processes: activation of membrane
receptors, signaling cascades and “core effectors” such as
metabolic pathways. Collected protein and small molecule
interaction data form the basis for building mechanistic
reconstruction for major processes in cell signaling and
metabolism and for making marked improvements in functional
ontologies. We describe tools and algorithms that allow
utilizing content of our database in the process of mining
different disease and drug-related “OMICs” datasets produced
in the context of drug discovery and other biomedical
research.
Elucidation of differential
response networks from gene expression data
Andrej Bugrim, Genego, Inc., USA
We describe a novel systems level approach to the analysis of gene expression data and the elucidation of biological networks affected by drug action. Specifically, some 15,000 human linear signaling and biochemical pathway modules were generated from “canonical” maps in MetaCore™ (GeneGo, Inc.), and used as templates for mapping microarray expression data from rat livers exposed to phenobarbital, mestranol and tamoxifen. We analyzed sample-to-sample distances, in gene expression space, of individual pathway modules in order to select “differential” pathways. These are defined by highly correlated expression among multiple repeats of the same treatment, and strong anti-correlation between different treatments. The gene content of these differential pathways is then used for generating network modules that distinguish between the treatments, followed by enrichment analysis using several ontologies. Unlike traditional techniques in microarray expression profiling, our method takes into account both network connectivity and gene expression profiles, and allows for the use of “whole genome” expression data, which is not restricted by fold change and p-value for individual data points. The method enables detection of important cellular mechanisms involved in drug response that would have been missed by traditional procedures of statistical and functional analysis.
Protein scoring based on
significance in biological networks
Andrej Bugrim, Genego, Inc., USA
While systems biology tools and approaches are gaining wide acceptance among molecular biologists and clinical researchers, two fundamental issues have emerged. The first one is how to use sets of available high-throughput molecular data to reconstruct biological networks that are truly relevant to the condition of interest. The second, even more important issue is how to utilize results of such reconstruction in the framework of standard laboratory practices and in clinical applications. We present a novel algorithm designed to evaluate importance of individual protein nodes for providing connectivity in condition-specific biological networks. The algorithm starts with a condition-specific set of genes or proteins (e.g. differentially expressed genes). First, we construct shortest path network connecting these genes using global database of interactions available in MetaCore™. Second, we evaluate the number of all paths traversing each node in the shortest path network in relation to the total number of paths going via the same node in the global network. Using these numbers as well as relative size of the initial gene set we calculate p-values for each node in the shortest path network, showing whether or not it is statistically significant for providing connectivity. We test algorithm’s ability to assign high significance to biologically validated drug targets by using public set of gene expression data from psoriatic patients. We show that our method is able to uncover many genes that do not show up on gene expression level but are nevertheless highly related to disease pathways. Then we proceed to demonstrate application of this approach in finding correlations between sets of genomic and proteomic data. The approach can be applied for uncovering new, higher-quality drug targets, validation of existing targets and cross-validation of genomic and proteomics or other types of data.
MetaTox: leveraging the power of
systems biology for improving drug safety
Andrej Bugrim, Genego, Inc., USA
It is now an accepted practice that evaluation of drug and chemical safety should involve understanding of perturbations caused by a compound in functional units of living cells – biological pathways, networks and modules. Current analytical procedures in toxicogenomics however are mainly focused on statistical analysis of expression patterns, aiming at identification of small sets of genes which are most characteristic for a certain treatment. MetaTox is the new concept that applies techniques of systems biology to predicting drug toxicity and understanding cellular mechanisms behind drug-response. In MetaTox we conduct analysis on the level of “functional descriptors”: pathways, network modules, and functional profiles, rather than analyzing individual genes. Such analysis allows building predictors that are multidimensional, robust, and allow mechanistic and functional interpretation. Moreover, these functional predictors enable one to consider drug safety in the context of specific indications. We demonstrate application of this concept to the analysis of real-life toxicogenomics datasets and describe MetaTox Consortium – a collaborative effort to improve drug safety evaluation by leveraging systems biology approaches.
Molecular simulations and
insights into molecular biology
Chandra Verma, Bioinformatics Institute, Singapore
Applications of mutli-resolution techniques in molecular
simulations are providing new insights into the functionings
of biological macromolecular machines and the regulations of
pathways and networks. These will be discussed using several
examples that have been developed in close association with
experiments.
Towards bridging the gap between
transcriptome and proteome measurements
Mahesan Niranjan, The University of Sheffield, UK
Much of the interesting cellular function in biology is
attributable to mechanisms that differentially regulate
concentrations of proteins.
High throughput measurements with microarrays capture the
profile of mRNA concentrations which are treated as proxies
for protein abundances. That there cannot be a one to one
correspondence between these two levels has been noted in
the literature by several authors.
Eukaryotic cells may employ a number of different mechanisms
to regulate protein levels. First and simplest is via a
direct link between the required protein levels and
transcription. The more protein is required, the greater is
the amount of transcribed mRNA.
Secondly, the mRNA can be degraded selectively. mRNA
molecules are usually unstable and the rate at which they
decay show a high variation governed by different
mechanisms. Thus controlling the decay rates selectively can
lead to different rates at which the corresponding protein
can be synthesised. Thirdly, ribosome binding can be
differential, achieving selectivity in the rate of
translation of protein molecules. Finally, there can be
differential regulation at the post translational level
whereby different proteins decay at different rates. Thus we
find very little correlation between mRNA concentrations and
the corresponding protein levels, as several authors have
noted.
Here I use a machine learning framework to predict protein
levels using mRNA levels and several proxies of mRNA level
regulation such as codon bias, transcript length, mass of
the resulting protein, measured ribosome occupancy and
measured mRNA halflives as input.
Using Gaussian process regression I show it is possible to
achieve a higher level of predictability of protein levels
than is apparent when one looks for correlation between mRNA
and protein abundances.
From this framework I show how a leave-one-out strategy can
be employed to uncover which of the mRNA-protein pairs do
not fit such data driven model based prediction. I argue
that these pairs that do not fit the learned model are
candidates for regulation at the post translational level,
i.e. failure, rather than success, of the model is
informative! Putting together a small dataset for yeast,
where simultaneous measurements of mRNA and protein levels
are available in the public domain, and the remaining
features can be extracted from databases and publications, I
ranked potential candidates for post translational
regulation. Searching through the literature it was possible
to find experimental evidence for five mRNA-prote in pairs
amongst the top twelve, confirming the hypothesis.
RNA - biology and secondary
structure
Peter Clote, Boston College, USA
According to the very recent ENCODE Consortium paper
appearing in Nature, the human genome is pervasively
transcribed; i.e. around 15% of the genome is transcribed
although only a fraction of the transcripts account for
mRNA, tRNA, rRNA, microRNA, etc. Is nature so wasteful as to
squander a large percent of the cell's energy resources
toward transcribing "junk RNA"? Or instead are we at the
very threshold of beginning to unravel the mystery of new
classes of RNA and their function? In this talk, we present
an overview of the chemistry and biology of RNA: glycosidic
bond, nonstandard RNAs, base stacking, Tinoco, free energy,
noncanonical base pairing, Leontis-Westhof classification,
3-dimensional motifs. We then discuss RNA secondary
structure, asymptotic results, dynamic programming, minimum
free energy structure prediction, and applications.
RNA - algorithms
Peter Clote, Boston College, USA
In this continuation, we discuss the Boltzmann partition
function for RNA secondary structure, sampling, applications
to siRNA design, as well as structural alignment algorithms
and some noncoding RNA gene finders.
Physics-based all-atom
modeling of protein dynamics
Yong Duan, University of California, USA
Physics-based modeling has become an indispensable tool
to study protein dynamics. Recent improvement in force field
and rapid increase in computer speed have made this method
increasingly powerful. Recent advances included the
successful simulations of a number of small proteins to
their native states with the structures to as close as
sub-angstrom from the experimental structures. These
exciting advances marked the beginning of accurate
simulations of protein folding to the native states of
proteins which have not been possible before. This talk is
divided into two parts.
In the first part, I will discuss the history and future of
AMBER force field and its applications to study protein
dynamics. AMBER, as one of the powerful simulation packages,
has been used in the biomolecular simulation community to
help tackle a wide-range of problems. The AMBER force
fields, first developed in 1981, have evolved into a
collection of force fields including the fully polarizable
version released in 2002. An exciting development is the
AMBER force field consortium which will steer the future
development of AMBER force fields. Two examples of the
applications of AMBER in the studies of protein dynamics
will be presented. In the first example, AMBER is used to
study the dynamics of nucleosome particles, to understand
the dynamics of the histone tails which are intimately
linked to gene control. In the second example, AMBER is used
to study G-protein Coupled Receptors which play key roles in
human physiology and are the targets of more than 40% of the
drugs.
In the second part, I will focus on the application of
physics-based simulations to study protein folding and
aggregation. Understanding the mechanisms of protein folding
is a key step to link protein primary sequences to their
structure and function and has been termed the second half
of genomics. The discovery of the folding-related diseases,
including the debilitating Alzheimer and Parkinson and other
neuro-degenerative diseases, underscores the need of a
comprehensive understanding of how proteins reach their
native states. We have attempted to understand how proteins
aggregate using small peptides as the model systems. In the
past, direct simulations of protein folding using
physics-based models have not been possible. Thanks to the
recent advances in force field, we have been able to reach
the native states of four small proteins. The simulations
unvealed a rather complex picture of protein folding.
Controllable gating of the
water permeation across nanoscale channels
Haiping Fang, Shanghai Institute of Applied Physics,
Chinese Academy of Sciences, China
In this talk, the dynamics of the single-file water
chains inside a single-walled carbon nanotube (SWNT) with an
appropriate radius was studied with molecular dynamics
simulations under the influence of continuous deformations
and/or external charges. It is found that the water
permeation across the channel has an excellent on-off gating
behaviour. The water conduction across the water channel
keeps almost fixed for a considerable deformation and/or a
very small distance of the external charge from the channel.
The channel closes rapidly when the deformation exceeds
and/or the distance of the external charge from the channel
is less than a threshold. We believe that this excellent
property is important for biological systems to achieve
accurate information transfer in an environment full of
thermal fluctuations and useful to develop SWNT-based
molecular machines.
Signal and motif detection in
genomic sequences
Jagath C. Rajapakse, Nanyang Technological
University, Singapore
Signals of genomics sequences refer to specific sites
relating to important biological phenomena, for example,
transcription start sites (TSS), translation initiation
sites (TIS), and splice sides (SS). The computational
techniques to detect these signals are becoming popular
because of the complexities and difficulties in determining
these sites experimentally. This tutorial will introduce
computational intelligence techniques, such as neural
networks, genetic algorithms, and their hybrids for the
detection of TSS, TIS, and SS.
Motifs in genomic sequences refer to short segments of DNA,
which are conserved and have some important biological
function. Most motifs in DNA sequences have regulatory
functions, for examples, transcription factor binding sites
(TFBS), promoters, and ribosome binding sites. Motif
detection is a difficult problem in computational biology
because the motif instances usually present in the sequences
with a considerable number of degenerations. We discuss
several approaches to motif detection, including profile
analysis, neural network methods, the MEME approach, etc. We
then proceed to discuss the graphical methods for week motif
detection problem where classical techniques fail.
Statistical physics of RNA
Ralf Bundschuh, Ohio State University, USA
Recap of Boltzmann partition function, molten RNA,
z-transforms, RNA denaturation, native-molten transition.
Quantitative modeling of
force-extension experiments
Ralf Bundschuh, Ohio State University, USA
Experimental techniques for force-extension experiments,
polymer physics of single-stranded RNA, secondary structure
in force-extension experiments, quantitative modeling,
structure determination through force-extension experiments.
RNA folding kinetics
Ralf Bundschuh, Ohio State University, USA
Types of experiments, modeling approaches, nanopore
experiments, modeling of nanopore experiments.
MicroRNA target prediction
Ralf Bundschuh, Ohio State University, USA
Biology of microRNAs, target prediction problem, overview
over target prediction software.
Physics-based all-atom modeling of
protein dynamics
Yong Duan, University of California, USA
Physics-based modeling has become an indispensable tool
to study protein dynamics. Recent improvement in force field
and rapid increase in computer speed have made this method
increasingly powerful. Recent advances included the
successful simulations of a number of small proteins to
their native states with the structures to as close as
sub-angstrom from the experimental structures. These
exciting advances marked the beginning of accurate
simulations of protein folding to the native states of
proteins which have not been possible before. This talk is
divided into two parts.
In the first part, I will discuss the history and future of
AMBER force field and its applications to study protein
dynamics. AMBER, as one of the powerful simulation packages,
has been used in the biomolecular simulation community to
help tackle a wide-range of problems. The AMBER force
fields, first developed in 1981, have evolved into a
collection of force fields including the fully polarizable
version released in 2002. An exciting development is the
AMBER force field consortium which will steer the future
development of AMBER force fields. Two examples of the
applications of AMBER in the studies of protein dynamics
will be presented. In the first example, AMBER is used to
study the dynamics of nucleosome particles, to understand
the dynamics of the histone tails which are intimately
linked to gene control. In the second example, AMBER is used
to study G-protein Coupled Receptors which play key roles in
human physiology and are the targets of more than 40% of the
drugs.
In the second part, I will focus on the application of
physics-based simulations to study protein folding and
aggregation. Understanding the mechanisms of protein folding
is a key step to link protein primary sequences to their
structure and function and has been termed the second half
of genomics. The discovery of the folding-related diseases,
including the debilitating Alzheimer and Parkinson and other
neuro-degenerative diseases, underscores the need of a
comprehensive understanding of how proteins reach their
native states. We have attempted to understand how proteins
aggregate using small peptides as the model systems. In the
past, direct simulations of protein folding using
physics-based models have not been possible. Thanks to the
recent advances in force field, we have been able to reach
the native states of four small proteins. The simulations
unvealed a rather complex picture of protein folding.
Algorithms for peptide sequencing
from tandem mass spectrometry
Hon Wai Leong, National University of Singapore
Tandem Mass spectrometry has become the technology of
choice in many proteomics projects. Computational analysis
of the MS/MS mass spectra generated by proteomics machines
have often been the bottleneck in applying this technology.
In this talk, we present a quick overview of computational
approaches for peptide sequencing.
We then focus on our work on peptide sequencing for multi
charge mass spectra. Most of the computational methods for
peptide sequencing from mass spectra have focused on singly
charged ion types. A few of these methods also consider
doubly charged ion types. However, there has been little
attention focused on multi charged
ion types (charge of 3 or more) even though multi charge
spectra (with charges up to 5) are publicly available, for
example, from the GPM web-site.
We present our recent work on using generalized model to
analyze and characterize multi charged mass spectra. Our
analysis shows that higher charge ions contributes
significantly to the specificity of the spectra. Our model
also allow us to derive upper bounds on sensitivity that
help to explain the relative poor performance of current
algorithms on these higher charge spectra.
Finally, we also present our initial work on de novo peptide
sequencing algorithms that are designed specifically for
multi charge mass spectra. Specifically, we present our
algorithm, called the GST-SPC algorithm that uses the
concept of strong tags as candidates for peptide extension
and use a graph theoretical algorithm using the generalized
spectrum graph model based on our multi-charge model. Our
experimental results show that our algorithm outperforms
existing algorithms on multi charge mass spectra. We also
present some current computational issues arising from this
research.
(This is joint work with Kang Ning, Ket Fah Chong and Nan Ye
of NUS and Pavel Pevzner of UC-San Diego.)
Sense-antisense human gene
expression pairs: data mining and analysis on global genome
scale
Vladimir Kuznetsov, Genome Institute of Singapore
Transcription of mRNAs from opposite strand to a given gene may cause numerous regulatory effects on gene expression, pathways and cellular functions. Computational approaches based on gene transcripts mapping onto human genome reported several thousands of naturally transcribed mRNAs from genes located on opposite strand of the same locus (cis-antisense (or sense-antisense) gene pairs (CASGP)). However, since reported databases use different sources of the sequence information (UniGene, EST, SAGE etc), they provide poorly compatible and essentially incomplete sets of sense-antisense (SA) gene pairs. To integrate the data on SA pair transcription we created united SA gene pairs database that map the latest GenBank RefSeq, mRNA and EST sequences onto human genome and re-map several previously published CASGP data sets (Y.L. Orlov, Jiangtao Zhou, V.A. Kuznetsov). Clustering of reported transcripts by chromosome coordinates revealed up to 9000 of SA loci. Analyzing our database, microarray expression datum and literature, we demonstrate that sense-antisense gene pairs can provide regulatory functions at several levels of gene expression process including alternative splicing, binding, translational regulation, RNA stability and trafficking. We also demonstrate the associations of different expression pattern of CASGP transcripts with phenotypes of different normal and cancer cells.
Probing the secondary structure
landscape of RNA
Peter Clote, Boston College, USA
In Nature 447(7146):799--816, 2007 the ENCODE Consortium
published a landmark paper which stated that the human
genome is "pervasively expressed"; indeed, while 14.7% of
genome is transcribed, only 1-2% of the transcript can be
accounted for by mRNA, rRNA, tRNA, miRNA, etc. The
intellectual end of the popular press reacted to the ENCODE
paper in the June 14, 2007 issue of Economist, which stated
“Molecular biology is undergoing its biggest shake-up in
50 years, as a hitherto little-regarded chemical called RNA
acquires an unsuspected significance.It is beginning to dawn
on biologists that they may have got it wrong. Not
completely wrong, but wrong enough to be embarrassing.â€
Put in a less castigating light, we can posit that the
ENCODE Consortium paper provides data underlines the
necessity of developing new algorithms to analyze RNA
secondary structures, evolutionarily related RNA sequences
and structures, and more generally to better understand the
landscape of RNA secondary structures. In this talk, we
present new algorithms developed by our lab to compute the
minimum free energy structure and partition function for
various classes of RNA, with potential application for
riboswitch detection. This work is joint with E. Freyhult,
J. Waldispuhl, B. Beshadi, and J.-M. Steyaert.
Some recent genomics and
structural biology web servers
Peter Clote, Boston College, USA
In this talk, we present an overview of various tools our lab has developed, mostly in the area of structural biology. We will discuss time warping and its use in functional genomics, disulfide connectivity, cysteine state prediction, beta-barrel transmembrane structure, and 3-dimensional motif detection in RNA.
This work is the collaboration of F. Ferre, W.A. Lorenz,
Y. Ponty, J. Waldispuhl and myself. Of particular
significance is the energy model and very deep use of
grammars in the transmembrane supersecondary structure
algorithm of Jerome Waldispuhl, now a lecturer at MIT.
