Bioinformatics: Applications
ZOO 4903
Fall 2006, MW 10:30-11:45
Sutton Hall, Room 312
Jonathan Wren
Systems Biology
Lecture overview
• What we’ve talked about so far
– Pathways & network motifs
– Simulating evolution in-silico
– Cellular simulations
• Overview
– The ultimate goal of biology & bioinformatics
is to tie it all together and understand the
system
– In the meantime, forced to live in the real
world, we focus on tying a few things
together
Systems Biology – backers
& attackers
Though coined 40 years ago, a lot of people still
ask, "What's that?" when the term systems
biology comes up. "It is used in so many different
contexts, nobody is really clear what you mean by
it," says John Yates III, a professor at the Scripps
Research Institute in La Jolla, Calif. He's not the
only one stumped by the term's meaning. David
Placek, president of Sausalito, Calif.-based Lexicon
Branding, a company that cooks up names for
pharmaceutical products such as Velcade and
Meridia, says he's not so hot on the moniker.
"Systems biology is just so general that it could
apply to many things. When you're naming a
category, the underlying principle is that if you
make a statement like, 'I'm doing systems
biology,' do people know what you're talking
about?'“……
Volume 17 | Issue 19 | 27
Oct. 6, 2003, The Scientist
What is “Systems Biology”?
Is this just another name for “physiology”?
The study of the mechanisms underlying complex
biological processes as integrated systems of many
interacting components. Systems biology involves (1)
collection of large sets of experimental data (2) proposal
of mathematical models that might account for at least
some significant aspects of this data set, (3) accurate
computer solution of the mathematical equations to
obtain numerical predictions, and (4) assessment of the
quality of the model by comparing numerical simulations
with the experimental data.
-(Leroy Hood, 1999)
Institute for Systems Biology
http://www.systemsbiology.org/
Why Systems Biology?
• On the technology side (PUSH): Capabilities for highthroughput data gathering that have made us aware
that biological networks have many more components
than we previously surmised.
• On the biology side (PULL): The realization that to the
extent that we don’t characterize biological systems
quantitatively in their full complexity, the scope and
accuracy of our understanding of those systems will be
compromised. (in classical experimental terms, the
uncontrolled variables in the system will undermine our
confidence in the conclusions we draw from our
experiments and observations)
Systems Biology vs. traditional cell and
molecular biology
• Experimental techniques in systems biology are
high throughput.
• Intensive computation is involved from the start
in systems biology, in order to organize the data
into usable computable databases.
• Exploration in traditional biology proceeds by
successive cycles of hypothesis formation and
testing; data accumulates during these cycles.
• Systems biology initially gathers data without
prior hypothesis formation; hypothesis formation
and testing comes during post-experiment data
analysis and modeling.
Genomics, Proteomics &
Systems Biology
Genomics
Proteomics
Systems Biology
1990
1995
2000
2005
2010
2015
2020
Modelling Tools
9
7
#
5
3
1
65-69
70-74
75-79
80-84
Period
85-89
90-94
95-99
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
BIOSSIM (1968)
ESSYN (1976)
SCAMP (1983)
SCOP (1986)
METAMOD (1986)
SIMFIT (1990)
METAMODEL (1991)
METASIM (1992)
KINSIM (1993)
GEPASI (1994)
METALGEN (1994 ?)
MIST (1995)
METABOLIKA (1997 ?)
METAFLUX (1997)
SIMFLUX (1997)
MNA (1998)
CELLMOD (1998)
FLUXMAP (1999)
METATOOL (1999)
VCELL (1999)
From Klaus Mauch, University of Stuttgart
Systems Biology is an
integration of data & approaches
Technologies to study
systems at different levels
• Genomics (HT-DNA sequencing)
• Mutation detection (SNP methods)
• Transcriptomics (Gene/Transcript
measurement, SAGE, gene chips,
microarrays)
• Proteomics (MS, 2D-PAGE, protein
chips, Yeast-2-hybrid, X-ray, NMR)
• Metabolomics (NMR, X-ray, capillary
electrophoresis)
Each system has methods for
modeling
Pi Calculus
Flux Balance Analysis
Petri Nets
Differential Eqs
Each system has methods for
modeling
Boolean Networks
Electrical Circuit Model
Cellular Automata
So how can we meaningfully
integrate the data?
System heterogeneity in size
& timescale
Atomic Scale
0.1 - 1.0 nm
Coordinate data
Dynamic data
0.1 - 10 ns
Molecular dynamics
Molecular Scale
1.0 - 10 nm
Interaction data
Kon, Koff, Kd
10 ns - 10 ms
Interactions
Cellular Scale
10 - 100 nm
Concentrations
Diffusion rates
10 ms - 1000 s
Fluid dynamics
System heterogeneity in size
& timescale
Tissue Scale
0.01m - 1.0 m
Metabolic input
Metabolic output
1 s – 1 hr
Process flow
Organism scale
0.01m – 4.0 m
Behaviors
Habitats
1 hr – 100 yrs
Mechanics
Ecosystem scale
1 km – 1000 km
Environmental impact
Nutrient flow
1 yr – 1000 yrs
Network Dynamics
Each of the scales does not
fit together seamlessly
• If one scale (e.g., protein-protein
interactions) behaves deterministically
and with isolated components, then we
can use plug-n-play approaches
• If it behaves chaotically or stochastically,
then we cannot
• Most biological systems lie between this
deterministic order and chaos: Complex
systems
Man-made Complex Devices
Intel Pentium 4
42 million transistors
Man-made Complex Devices
• The Intel Itanium 2
• 410 million transistors
• Number of gates > 100 Million
By 2007 both Intel and AMD
are predicting dies with 1
billion transistors
In terms of parts and
interconnections, man-made
devices will likely have
comparable complexity to
bacterial cells if not greater by
around 2010
System Models
Building computational models of systems seems more
and more like a viable project.
Such a project would bring a much clearer understanding
of how systems are controlled and ultimately it should
bring unprecedented predictive power.
Are Biologists Ready?
Xo
S1
S2
S3
v
S4
S5
S6
X1
Xo and X1 fixed,
all reactions
reversible, assume
stable steady state.
Are Biologists Ready?
50 %
Xo
S1
S2
S3
v
S4
S5
S6
X1
What happens to the steady state?
Xo and X1 fixed,
all reactions
reversible, assume
stable steady state.
Are Biologists Ready?
50 %
Xo
S1
S2
S3
v
S4
S5
Typical replies:
1. Nothing happens.
2. Nothing happens unless it is the rate-limiting step.
3. The rate v goes down, but that’s all.
4. S3 goes up.
5. S4 goes down.
6. Species downstream of v go down.
7. Steady State flow changes but species levels don’t.
8. Xo and X1 change
S6
X1
Are Biologists Ready?
50 %
Xo
S1
S2
S3
v
S4
S5
S6
If we can’t understand this system how can we hope to understand:
X1
Functional Motif
Identification
Computer simulation of EGF
signal transduction PC12 cells.
Frances Brightman, Simon Thomas and David
Fell
http://bms-mudshark.brookes.ac.uk/frances/fabweb5.htm
29 species
Functional Motif
Identification
Computer simulation of EGF
signal transduction PC12 cells.
Frances Brightman, Simon Thomas and David
Fell
http://bms-mudshark.brookes.ac.uk/frances/fabweb5.htm
29 species
Functional Motif Identification
27 components
Functional Motif Identification
As we begin to connect systems
we can engage in inference
• We move up the chain from data to
knowledge by questioning, observing and
then hypothesizing
– These X genes are upregulated together, but
are they interacting?
– PPI network data suggests Y are
– Are these Y part of a complex?
– If they are always expressed together, that
suggests maybe yes
• As more data is integrated and systems
linked together, this becomes easier
Example of inference
(a) An interaction network of Snz–Sno proteins of S.
cerevisiae. The nodes represent proteins and the
lines represent yeast two-hybrid (Y2H) interactions.
The red nodes represent proteins that correspond to
genes in one transcriptome cluster, whereas the
green nodes represent proteins that correspond to
genes belonging to a different cluster. The existence
of two stable complexes can be hypothesized based
on the integrated data.
(b) The genes NTH1 and YLR270W have similar
expression profiles (upper panel). Red indicates
upregulation and green indicates downregulation.
mRNA expressions of both genes are upregulated
during heat shock and other forms of stress.
Deletions of NTH1 and YLR270W each confer
similar heat-shock sensitive phenotypes (lower
panel).
Integrating heterogeneous but related
observations
How are the data related?
What kind of model?
What kind of inferencing?
Is the data validated?
Can we take a “best guess” on how it
might work by drawing upon other
motifs or systems with similar
properties?
Problems?
How is static data interpreted since it’s a dynamic
system?
How do we deal with low-resolution quality?
How do we treat missing data?
How do we deal with heterogeneous data types?
How can we identify and evaluate competing
hypotheses inferred by any system?
Yes…
SB is springing out of
existing efforts anyway
• E-cell (Keio University, Japan)
• BioSpice Project (Arkin, Berkeley)
• Metabolic Engineering Working Group (Palsson
& Church, UCSD, Harvard)
• Silicon Cell Project (Netherlands)
• Virtual Cell Project (UConn)
• Gene Network Sciences Inc. (Cornell)
• Project CyberCell (Edmonton/Calgary)
So where do we start?
• Quantitative analysis
of components and
dynamics of complex
biological systems
Static (Tier 1)
Deterministic (Tier 2)
Stochastic (Tier 3)
Features of complex systems
• Nonlinearity
global properties not simple sum of parts
Features of complex systems
• Feedback loops
Features of complex systems
• Open systems (dissipation of energy)
Flagella uses energy:
Features of complex systems
• Can have memory (response history
dependent)
New protein may remain in
cell after initial response,
shifting the rate of reaction
the next time the cell is
exposed to a chemical
Response
Chemical concentration
Features of complex systems
• Nested (modules have complexity)
Features of complex systems
• There are no precise boundaries
So where do we start?
• Quantitatively account
for these properties
Static (Tier 1)
– Different levels of
modeling
• Three tiers
Deterministic (Tier 2)
– Static interactions
– Deterministic
– Stochastic
• Principles which
transcend tiers…
Stochastic (Tier 3)
Principle 1: Modularity
• Module
– Interacting nodes w/
common function
– Constrained pleiotropy
– Feedback loops,
oscillators, amplifiers
Principle 2: Recurring circuit
elements
• Network motifs
– Common methods to achieve an effect
Principle 3: Robustness
• Robustness
– Insensitivity to
parameter variation
• Severe constraints
on design
– Robustness not
present in most
designs
Aims of systems biology
• Tier 1: Interactome
– Which molecules talk
to each other in
networks?
• Tier 2: Deterministic
– What is the average
case behavior?
• Tier 3: Stochastic
– What is the variance
of the system?
Aims of systems biology
• Tier 1
– Get parts list
Aims of systems biology
• Tier 2 & 3
– Enumerate
biochemistry
– Define
network/mathematical
relationships
– Compute numerical
solutions
Aims of systems biology
• Tier 2 & 3
– Deterministic: Behavior of
system with respect to
time is predicted with
certainty given initial
conditions
– Stochastic: Dynamics
cannot be predicted with
certainty given initial
conditions
Aims of systems biology
• Deterministic
– Ordinary differential
equations (ODE’s)
• Concentration as a
function of time only
– Partial differential
equations (PDE’s)
• Concentration as a
function of space and
time
• Stochastic
– Stochastic update
equations
• Molecule numbers as
random variables
• functions of time
Y = # molecules at time t
Tier 1: Static interactome
analysis
• Protein-protein
– Signal transduction
– Cell cycle
• Protein-DNA
– Gene regulation
• Metabolic pathways
– Respiration
– cAMP
Tier 1: Static interactome
analysis
• Goals
– Determine network topology
– Network statistics
– Analyze modular structure
Tier 1: Static interactome
analysis
• Limitations:
– Time, space,
population
average
– Crude interactions
typical interactome
• strength
• types
– Global features
• starting point for
Tier 2 & 3
first time-varying yeast interactome (Bork 2005)
Tier 1: Static interactome analysis
• Analysis methods
– Functional
Genomics
• expression analysis
• network integration
– Graph Theory
• scale free
• small world
Tier 2: Deterministic Models
• Goal
– model mesoscale
system
– average case
behavior
lumped cell
• Three levels
– ODE system
– ODE compartment
system
– PDE
– data limited…
cell compartments
continuous time & space
(MinCDE oscillation)
Tier 2: Deterministic
Modeling
• Results
– Robust Chemotaxis
(Barkai 1997)
– MinCDE Oscillation
(Howard 2003)
– Feedback in Signal
Transduction (Brandman
2005)
• Output
– time series plots (ODE)
– condition on parameter
values
Brandman 2005
Tier 2: Deterministic Modeling
• Example
– Robustness in bacterial
chemotaxis
• Bacterial chemotaxis
robust to parameter
fluctuations!
– Chemotaxis: bacterial
migration towards/away
from chemicals
– Parameters
• concentrations
• binding affinities
Tier 2: Deterministic Modeling
• Bacterial chemotaxis
– model as random walk
• Exact adaptation
– change in concentration
of chemical stimulant
– rapid change in bacterial
tumbling frequency…
– then adapts back
precisely to its prestimulus value!!
Random walk
Experimental Design
• Is exact adaptation robust to substantial
variations in biochemical parameters?
• Systematically varied concentrations of
chemotaxis-network proteins and measured
resulting behavior
Distinguish between robust-adaptation
and fine-tuned models of chemotaxis
Tumbling frequency
IPTG inducer
pUA4
pUA4
E. Coli cheR -/- population
pUA4
Adaption time
pUA4
Express CheR over a
100-fold range
Adaption precision
1 mM L-aspartate
Adaptation precision = ratio of steady-state tumbling frequency
of unstimulated to stimulated cells
Summary of results
Tumbling frequency
0.3 ± 0.06 (20-fold)
Adaption time
3 ± 1 (3-fold)
Adaption precision
1.04 ± 0.07
Tumbling frequency as a function of time for wild-type cells
Conclusions from study
• Exact adaptation is maintained
despite substantial varations in
network-protein concentrations
– Exact adaptation is a robust
property
– …but adaptation time and steadystate behavior are fine-tuned
Tier 3: Stochastic analysis
• Fluctuations in abundance of expressed
molecules at the single-cell level
– Leads to non-genetic individuality of isogenic
population
Tier 3: Stochastic Analysis
• When stochasticity is
negligible, use
deterministic modeling…
• Molecular “noise” is
low:
– System is large
• molar quantities
– Fast kinetics
• reaction time negligible
– Large cell volume
• infinite boundary
conditions
Tier 3: Stochastic Analysis
• Molecular “noise” is high:
– System is small
• finite molecule count
matters
– Slow kinetics
• relative to movement time
– Large cell volume
• relative to molecule size
• Need explicit stochastic
modeling!
Tier 3: Ensemble Noise
• Transcriptional bursting
– Leaky transcription
– Slow transitions between chromatin states
• Translational bursting
– Low mRNA copy number
Tier 3: Temporal Noise
Canonical way of modeling molecular stochasticity
Tier 3: Spatial Noise
Finite number effect: translocation of molecules from the nucleus
to the cytoplasm have a large effect on nuclear concentration
Cytoplasm
N = average molecular abundance
η (coefficient of variation) = σ/N
• Decrease in abundance results ina
1/√N scaling of the noise (η=1/√N)
Nucleus
Recap
• Three tiers
– Interactomes
– Deterministic
– Stochastic
Static (Tier 1)
• Principles which cross tiers
– Modularity
– Reuse
– Robustness
Deterministic (Tier 2)
Stochastic (Tier 3)
Major challenges and limitations
• Measurement of chemical kinetics
parameters and molecular
concentrations in vivo
– Differences between in vitro and in vivo
data
• Compartmental specific reactions
Major challenges and limitations
• Data is the limit!!!
– Functional genomic data
(Interactomes)
– E. Coli chemotaxis (Leibler,
deterministic/robustness)
• Important
– parameter estimation
– feedback based estimation
methods
Sachs 2005
Software
• Tier 1: Interactomes
– Graphviz, Bioconductor, Cytoscape
• Tier 2: Deterministic
– Matlab (SBtoolbox), Mathematica
(PathwayLab)
• Tier 3: Stochastic
– R, Stochsim
Software
• High-performance algorithms to solve
systems of PDE’s
– Virtual Cell
• Automated parsing of networks into
stochastic and deterministic regimes
– H-GENESIS
– STOCK
Summary
• Systems Biology can be done by
breaking down each system into
modules
• Many problems remain unsolved in
exactly how to do this, but independent
efforts are being developed in most
areas that may one day merge together
For next time
• Read supplemental material S9
• Homework #10 due