synthetic biology - modeling and optimisation
DESCRIPTION
This tutorial was given at GECCO 2009. It updates and extends the slideshow "Executable Biology Tutorial"TRANSCRIPT
/203
Natalio KrasnogorASAP - Interdisciplinary Optimisation LaboratorySchool of Computer Science
Centre for Integrative Systems BiologySchool of Biology
Centre for Healthcare Associated InfectionsInstitute of Infection, Immunity & Inflammation
University of Nottingham
Synthetic Biology: Modelling and Optimisation
Copyright is held by the author/owner(s).
GECCO’09, July 8–12, 2009, Montréal Québec, Canada.
ACM 978-1-60558-505-5/09/07.
1
Thursday, 9 July 2009
/203
Outline•Brief Introduction to Computational Modeling
•Modeling for Top Down SB•Executable Biology•A pinch of Model Checking
•Modeling for the Bottom Up SB•Dissipative Particle Dynamics
•Automated Model Synthesis and Optimisation
•Conclusions2
Thursday, 9 July 2009
/203
Outline•Brief Introduction to Computational Modeling
•Modeling for Top Down SB•Executable Biology•A pinch of Model Checking
•Modeling for the Bottom Up SB•Dissipative Particle Dynamics
•Automated Model Synthesis and Optimisation
•Conclusions3
Thursday, 9 July 2009
/203
Synthetic Biology• Aims at designing, constructing and developing artificial biological systems •Offers new routes to ‘genetically modified’ organisms, synthetic living entities, smart drugs and hybrid computational-biological devices.
• Potentially enormous societal impact, e.g., healthcare, environmental protection and remediation, etc
• Synthetic Biology's basic assumption:• Methods commonly used to build non-biological systems could also be use to specify, design, implement, verify, test and deploy novel synthetic biosystems. • These method come from computer science, engineering and maths.• Modelling and optimisation run through all of the above.
4
Thursday, 9 July 2009
/203
Models and Reality
•The use of models is intrinsic to any scientific activity.
•Models are abstractions of the real-world that highlight some key features while ignoring others that are assumed to be not relevant.
•A model should not be seen or presented as representations of the truth, but instead as a statement of our current knowledge.5
Thursday, 9 July 2009
/203
What is modelling?
• Is an attempt at describing in a precise way an understanding of the elements of a system of interest, their states and interactions
• A model should be operational, i.e. it should be formal, detailed and “runnable” or “executable”.
6
Thursday, 9 July 2009
/203
•“feature selection” is the first issue one must confront when building a model
•One starts from a system of interest and then a decision should be taken as to what will the model include/leave out
•That is, at what level the model will be built
7
Thursday, 9 July 2009
/203
The goals of Modelling
•To capture the essential features of a biological entity/phenomenon•To disambiguate the understanding behind those features and their interactions•To move from qualitative knowledge towards quantitative knowledge
8
Thursday, 9 July 2009
/203
•There is potentially a distinction between modelling for Synthetic Biology and Systems Biology:
•Systems Biology is concerned with Biology as it is•Synthetic Biology is concerned with Biology as it could be
“Our view of engineering biology focuses on the abstraction and standardization of biological components” by R. Rettberg @ MIT newsbite August 2006.“Well-characterized components help lower the barriers to modelling. The use of control elements (such as temperature for a temperature-sensitive protein, or an exogenous small molecule affecting a reaction) helps model validation” by Di Ventura et al, Nature, 2006
9
Thursday, 9 July 2009
/203
•There is potentially a distinction between modelling for Synthetic Biology and Systems Biology:
•Systems Biology is concerned with Biology as it is•Synthetic Biology is concerned with Biology as it could be
“Our view of engineering biology focuses on the abstraction and standardization of biological components” by R. Rettberg @ MIT newsbite August 2006.“Well-characterized components help lower the barriers to modelling. The use of control elements (such as temperature for a temperature-sensitive protein, or an exogenous small molecule affecting a reaction) helps model validation” by Di Ventura et al, Nature, 2006
Co-design of parts and their models hence improving and making both more reliable
9
Thursday, 9 July 2009
/203
Cells
DNA/RNA
OrganCell colony
Proteins
Individual
RegulatoryNetworks
Thus, Multi-Scale Modelling in the 2 SBs seek to produce computable understanding integrating massive datasets at various levels of details simultaneously
Time
Progress
10
Thursday, 9 July 2009
/203
The Pragmalogical Problem of Modelling in XXI century Biology
• XXI century Biology brings to the fore the ubiquitous philosophical questions in complex systems, that of emergent behavior and the tension between reductionism and holistic approaches to science.
• Synthetic Biology (and SysBio) has, however, a very pragmatic agenda: the engineering and control of novel biological systems
• The pragmalogical problem: If each subcomponent of a living system (and processes/components therein) are understood… Can we say that the system is understood? That is, can we assume that the system = ∑parts ?
• More importantly: can we control that biosystem?
11
Thursday, 9 July 2009
/203
The Pragmalogical Problem of Modelling in XXI century Biology
• XXI century Biology brings to the fore the ubiquitous philosophical questions in complex systems, that of emergent behavior and the tension between reductionism and holistic approaches to science.
• Synthetic Biology (and SysBio) has, however, a very pragmatic agenda: the engineering and control of novel biological systems
• The pragmalogical problem: If each subcomponent of a living system (and processes/components therein) are understood… Can we say that the system is understood? That is, can we assume that the system = ∑parts ?
• More importantly: can we control that biosystem?
& Integrative
11
Thursday, 9 July 2009
/203
Modelling relies on rigorous computational, engineering and mathematical tools & techniques
However, the act of modelling remains at the interface between art and science
Undoubtedly, a multidisciplinary endeavour
12
Thursday, 9 July 2009
/203
Modelling as a constrained scientific art
Although modelling lies at the interface of art and science there are guidelines we can follow
Some examples: The scale separation map [Hoekstra et al, LNCS 4487, 2007]
Tools suitability & cost [Goldberg, 2002]
13
Thursday, 9 July 2009
/203
The Scale Separation Map
The Scale Separation Map is an abstraction recently proposed by Hoekstra and co-workers [Hoekstra et al, LNCS 4487, 2007]
Introduced in the context of Multi-scale modelling with cellular automata but the core concepts still valid for other modelling techniques
14
Thursday, 9 July 2009
/203
A Cellular Automata is defined as:C= < A(Δx, Δt,L,T), S, R, G, F >
A is a spatial domain made of cells of size Δx with a total size of LThe simulation clock ticks every Δt units for a total of T units
We can simulate processes: as fast as Δt for as long as T units ranging from Δx to L sizes.
The Scale Separation Map
Δx
L
L
ΔtT
15
Thursday, 9 July 2009
/203
A Scale Separation Map (SSM) is a two dimensional map with horizontal axis representing time and vertical axis representing space
Temporal scale (log)
Spa
tial s
cale
(log
)
1
23.1 3.2
0
A ξA
τA
B ξB
τB
16
Thursday, 9 July 2009
/203
A Scale Separation Map (SSM) is a two dimensional map with horizontal axis representing time and vertical axis representing space
Temporal scale (log)
Spa
tial s
cale
(log
)
1
23.1 3.2
0
A ξA
τAB ξB
τB
• Region 0: A and B overlap single scale multi-science model• Region 1: ξA ≈ ξB ^ τA > τB
temporal scale separation• Region 2: ξA > ξB ^ τB ≈ τA coarse and fine structures in similar timescales• Region 3.1: ξA > ξB ^ τB < τA familiar micro-macro models• Region 3.2: ξA > ξB ^ τB > τA small and slow process linked to a fast and large process (e.g. Blood flood and artery repair)
16
Thursday, 9 July 2009
/203
A Scale Separation Map (SSM) is a two dimensional map with horizontal axis representing time and vertical axis representing space
Temporal scale (log)
Spa
tial s
cale
(log
)
1
23.1 3.2
0
A ξA
τAB ξB
τB
• Region 0: A and B overlap single scale multi-science model• Region 1: ξA ≈ ξB ^ τA > τB
temporal scale separation• Region 2: ξA > ξB ^ τB ≈ τA coarse and fine structures in similar timescales• Region 3.1: ξA > ξB ^ τB < τA familiar micro-macro models• Region 3.2: ξA > ξB ^ τB > τA small and slow process linked to a fast and large process (e.g. Blood flood and artery repair)
16
Thursday, 9 July 2009
/203
A Scale Separation Map (SSM) is a two dimensional map with horizontal axis representing time and vertical axis representing space
Temporal scale (log)
Spa
tial s
cale
(log
)
1
23.1 3.2
0
A ξA
τA
B ξB
τB
• Region 0: A and B overlap single scale multi-science model• Region 1: ξA ≈ ξB ^ τA > τB
temporal scale separation• Region 2: ξA > ξB ^ τB ≈ τA coarse and fine structures in similar timescales• Region 3.1: ξA > ξB ^ τB < τA familiar micro-macro models• Region 3.2: ξA > ξB ^ τB > τA small and slow process linked to a fast and large process (e.g. Blood flood and artery repair)
16
Thursday, 9 July 2009
/203
A Scale Separation Map (SSM) is a two dimensional map with horizontal axis representing time and vertical axis representing space
Temporal scale (log)
Spa
tial s
cale
(log
)
1
23.1 3.2
0
A ξA
τA
B ξB
τB
• Region 0: A and B overlap single scale multi-science model• Region 1: ξA ≈ ξB ^ τA > τB
temporal scale separation• Region 2: ξA > ξB ^ τB ≈ τA coarse and fine structures in similar timescales• Region 3.1: ξA > ξB ^ τB < τA familiar micro-macro models• Region 3.2: ξA > ξB ^ τB > τA small and slow process linked to a fast and large process (e.g. Blood flood and artery repair)
16
Thursday, 9 July 2009
/203
A Scale Separation Map (SSM) is a two dimensional map with horizontal axis representing time and vertical axis representing space
Temporal scale (log)
Spa
tial s
cale
(log
)
1
23.1 3.2
0
A ξA
τA
B ξB
τB
• Region 0: A and B overlap single scale multi-science model• Region 1: ξA ≈ ξB ^ τA > τB
temporal scale separation• Region 2: ξA > ξB ^ τB ≈ τA coarse and fine structures in similar timescales• Region 3.1: ξA > ξB ^ τB < τA familiar micro-macro models• Region 3.2: ξA > ξB ^ τB > τA small and slow process linked to a fast and large process (e.g. Blood flood and artery repair)
16
Thursday, 9 July 2009
/203
Even within a single cell the space & time scale separations are important
[F.J. Romero Campero, 2007]
E.g.:
• Within a cell the dissociation constants of DNA/ transcription factor binding to specific/non-specific sites differ by 4-6 orders of magnitude
• DNA protein binding occurs at 1-10s time scale very fast in comparison to a cell’s life cycle.
17
Thursday, 9 July 2009
/203
Temporal scale (log)
Spa
tial s
cale
(log
)
• With sufficient data each process can be assigned its space-time region unambiguously
• A given process may well have its Δx (respectively Δt) > than another’s ξA (respectively τA)
• Hence different processes in the SSM might require different modelling techniques
Couplings, e.g. F
The Scale Separation Map
18
Thursday, 9 July 2009
/203
Modelling Approaches
There exist many modelling approaches, each with its advantages and disadvantages.
Macroscopic, Microscopic and MesoscopicQuantitative and qualitativeDiscrete and ContinuousDeterministic and StochasticTop-down or Bottom-up
19
Thursday, 9 July 2009
/203
Modelling Frameworks
•Denotational Semantics Models:
Set of equations showing relationships between molecular quantities and how they change over time.They are approximated numerically. (I.e. Ordinary Differential Equations, PDEs, etc)
•Operational Semantics Models:
Algorithm (list of instructions) executable by an abstract machine whose computation resembles the behaviour of the system under study. (i.e. Finite State Machine)
Jasmin Fisher and Thomas Henzinger. Executable cell biology. Nature Biotechnology, 25, 11, 1239-1249 (2008)
20
Thursday, 9 July 2009
/203
From [D.E Goldberg, 2002] (adapted): “Since science and math are in the description
business, the model is the thing…The engineer or inventor has much different motives. The engineered object is the thing”
ε, e
rror
C, cost of modelling
Synthetic Biologist
Computer Scientist/Mathematician
Tools Suitability and Cost
21
Thursday, 9 July 2009
/203
Tools Suitability and Cost
Low cost/High error
High cost/Low error
Unarticulatedwisdom
ArticulatedQualitativemodels
Dimensionalmodels
Facetwisemodels
EquationsOf motion
ChemicalMarkup Language
(CML)
BioinformaticSequence MarkupLanguage (BSML)
BiopolimerMarkup Language
(BioML)
Microarrays and G.E. Markup Language
(MAGEML)
Cell Markup Language
(MathML)
Systems Biology Markup Language
(SBML)
Mathematics Markup Language
(MathML)
Adapted from [Goldberg 2002]
22
Thursday, 9 July 2009
/203
Formalism-independent errors Formalism-dependent errors
Low cost/High error
High cost/Low error
Unarticulatedwisdom
Dimensionalmodels
Facetwisemodels
EquationsOf motion
From [Di Ventura et al., Nature, 2006]
23
Thursday, 9 July 2009
/203
Formalism-independent errors Formalism-dependent errors
Low cost/High error
High cost/Low error
Unarticulatedwisdom
Dimensionalmodels
Facetwisemodels
EquationsOf motion
From [Di Ventura et al., Nature, 2006]
23
Thursday, 9 July 2009
/203
Formalism-independent errors Formalism-dependent errors
Low cost/High error
High cost/Low error
Unarticulatedwisdom
Dimensionalmodels
Facetwisemodels
EquationsOf motion
From [Di Ventura et al., Nature, 2006]
23
Thursday, 9 July 2009
/20324
Thursday, 9 July 2009
/203
Stochasticity in Cellular Systems Most commonly recognised sources of noise in cellular system are low
number of molecules and slow molecular interactions.
Over 80% of genes in E. coli express fewer than a hundred proteins per cell.
Mesoscopic, discrete and stochastic approaches are more suitable: Only relevant molecules are taken into account. Focus on the statistics of the molecular interactions and how often they
take place.
Mads Karn et al. Stochasticity in Gene Expression: From Theories to Phenotypes. Nature Reviews, 6, 451-464 (2005)
Purnananda Guptasarma. Does replication-induced transcription regulate synthesis of the myriad low copy number poteins of E. Coli. BioEssays, 17, 11, 987-997
25
Thursday, 9 July 2009
/203
“Although the road ahead is long and winding, it leads to a future where biology and medicine are transformed into precision engineering.” - Hiroaki Kitano.
Synthetic Biology and Systems biology promise more than integrated understanding: it promises systematic control of biological systems:
1. From an experimental viewpoint: Improved data acquisition 2. From a bioinformatics viewpoint: Improved data analysis tools 3. From a conceptual viewpoint: move from a science of mass-action/
energy-conversion to a science of information processing through multiple heterogeneous medium
Towards Executable Modells for SBs
26
Thursday, 9 July 2009
/203
There are good reasons to think that information processing is a key viewpoint to take when modeling
Life as we know is:• coded in discrete units (DNA, RNA, Proteins)• combinatorially assembles interactions (DNA-RNA, DNA-Proteins,RNA-Proteins , etc) through evolution and self-organisation• Life emerges from these interacting parts• Information is:
• transported in time (heredity, memory e.g. neural, immune system, etc)• transported in space (molecular transport processes, channels, pumps, etc)
• Transport in time = storage/memory a computational process• Transport in space = communication a computational process• Signal Transduction = processing a computational process
27
Thursday, 9 July 2009
/203
It thus makes sense to use methodologies designed to cope with complex, concurrent, interactive systems of parts as found in computer sciences (e.g.): Petri Nets Process Calculi P-Systems
28
Thursday, 9 July 2009
/203
InfoBioticswww.infobiotic.net
•The utilisation of cutting-edge information processing techniques for biological modelling and synthesis•The understanding of life itself as multi-scale (Spatial/Temporal) information processing systems•Composed of 3 key components:•Executable Biology (or other modeling techniques)•Automated Model and Parameter Estimation•Model Checking (and other formal analysis)
29
Thursday, 9 July 2009
/203
Cells
Colonies
Modeling in Systems & Synthetic Biology
Networks
Systems Biology Synthetic Biology
• Understanding• Integration• Prediction• Life as it is
•Control• Design• Engineering•Life as it could be
Computational modelling toelucidate and characterisemodular patterns exhibitingrobustness, signal filtering,amplification, adaption, error correction, etc.
Computational modelling toengineer and evaluate possible cellular designsexhibiting a desiredbehaviour by combining well studied and characterised cellular modules
30
Thursday, 9 July 2009
/203
• It is a hard process to design suitable models in systems/synthetic biology where one has to consider the choice of the model structure and model parameters at different points repeatedly. • Some use of computer simulation has been mainly focused on the computation of the corresponding dynamics for a given model structure and model parameters.
• Ultimate goal: for a new biological system (spec) one would like to estimate the model structure and model parameters (that match reality/constructible) simultaneously and automatically.
• Models should be clear & understandable to the biologist
Model Design in Systems/Synthetic Biology
31
Thursday, 9 July 2009
/203
How you select features, disambiguate and quantify depends on the goals behind your modelling enterprise.
Sys
tem
s B
iolo
gy
Syn
thet
ic B
iolo
gy
Basic goal: to clarify current understandings by formalising what the constitutive elements of a system are and how they interact
Intermediate goal: to test current understandings against experimental data
Advanced goal: to predict beyond current understanding and available data
Dream goal: (1) to combinatorially combine in silico well-understood
components/models for the design and generation of novel experiments and hypothesis and ultimately
(2) to design, program, optimise & control (new) biological systems
32
Thursday, 9 July 2009
/203
Model Development From [E. Klipp et al, Systems Biology in Practice,
2005]1. Formulation of the problem2. Verification of available information3. Selection of model structure4. Establishing a simple model5. Sensitivity analysis6. Experimental tests of the model predictions7. Stating the agreements and divergences between
experimental and modelling results8. Iterative refinement of model
33
Thursday, 9 July 2009
/203
Outline•Brief Introduction to Computational Modeling
•Modeling for Top Down SB•Executable Biology•A pinch of Model Checking
•Modeling for the Bottom Up SB•Dissipative Particle Dynamics
•Automated Model Synthesis and Optimisation
•Conclusions34
Thursday, 9 July 2009
/203
Executable Biology with P systems
Field of membrane computing initiated by Gheorghe Păun in 2000
Inspired by the hierarchical membrane structure of eukaryotic cells
A formal language: precisely defined and machine processable
An executable biology methodology
35
Thursday, 9 July 2009
/203
Functional EntitiesContainer
• A boundary defining self/non-self (symmetry breaking).• Maintain concentration gradients and avoid environmental damage.
Metabolism
• Confining raw materials to be processed.• Maintenance of internal structures (autopoiesis).
Information
• Sensing environmental signals / release of signals.• Genetic information
36
Thursday, 9 July 2009
/203
Distributed and parallel rewritting systems in compartmentalised hierarchical structures.
Compartments
Objects
Rewriting Rules
• Computational universality and efficiency.
• Modelling Framework
37
Thursday, 9 July 2009
/203
Cell-like P systems
the classic P system diagram appearing in most papers (Păun)
Intuitive Visual representation as a Venn diagram with a unique superset and without intersected sets.
38
Thursday, 9 July 2009
/203
Cell-like P systems
1
23
4
75
8
6
9
formally equivalent to a tree:
the classic P system diagram appearing in most papers (Păun)
Intuitive Visual representation as a Venn diagram with a unique superset and without intersected sets.
38
Thursday, 9 July 2009
/203
Cell-like P systems
• a string of matching parentheses: [1 [2 ]2 [3 ]3 [4 [5 ]5 [6 [8 ]8 [9 ]9 ]6
[7 ]7 ]4 ]1
1
23
4
75
8
6
9
formally equivalent to a tree:
the classic P system diagram appearing in most papers (Păun)
Intuitive Visual representation as a Venn diagram with a unique superset and without intersected sets.
38
Thursday, 9 July 2009
/203
P-Systems: Modelling PrinciplesMoleculesStructured Molecules
ObjectsStrings
Molecular Species Multisets of objects/strings
Membranes/organelles Membrane
Biochemical activity rules
Biochemical transport Communication rules
39
Thursday, 9 July 2009
/203
Stochastic P Systems
40
Thursday, 9 July 2009
/203
Rewriting Rules
used by Multi-volume Gillespie’s algorithm
41
Thursday, 9 July 2009
/203
Molecular Species A molecular species can be represented using
individual objects.
A molecular species with relevant internal structure can be represented using a string.
42
Thursday, 9 July 2009
/203
Molecular Interactions Comprehensive and relevant rule-based schema
for the most common molecular interactions taking place in living cells.
Transformation/Degradation Complex Formation and Dissociation Diffusion in / out Binding and Debinding Recruitment and Releasing Transcription Factor Binding/Debinding Transcription/Translation
43
Thursday, 9 July 2009
/203
Compartments / Cells Compartments and regions are explicitly
specified using membrane structures.
44
Thursday, 9 July 2009
/203
Colonies / Tissues Colonies and tissues are representing as
collection of P systems distributed over a lattice.
Objects can travel around the lattice through translocation rules.
v
45
Thursday, 9 July 2009
/203
Molecular Interactions Inside Compartments
46
Thursday, 9 July 2009
/203
Passive Diffusion of Molecules
47
Thursday, 9 July 2009
/20348
Thursday, 9 July 2009
/203
a b
a b
Symport channel
a
a
b
b
Antiport channel
ab
a b
Promoted symport channel (trap)
c
Transport Modalities
49
Thursday, 9 July 2009
/203
1
5
4 3
2
Endocitosys
Pinocitosys
Phagocitosys
Exocitosys
Transport Modalities
50
Thursday, 9 July 2009
/203
Transport Modalities
Highly specific:cell specific & topology specific
51
Thursday, 9 July 2009
/203
Signal Sensing and Active Transport
52
Thursday, 9 July 2009
/203
Specification of Transcriptional Regulatory Networks
53
Thursday, 9 July 2009
/203
Transcription as Rewriting Rules on Multisets of Objects and Strings
54
Thursday, 9 July 2009
/203
Translation as Rewriting Rules on Multisets of Objects and Strings
55
Thursday, 9 July 2009
/203
Post-Transcriptional Processes For each protein in the system, post-transcriptional processes like
translational initiation, messenger and protein degradation, protein dimerisation, signal sensing, signal diffusion etc are represented using modules of rules.
Modules can have also as parameters the stochastic kinetic constants associated with the corresponding rules in order to allow us to explore possible mutations in the promoters and ribosome binding sites in order to optimise the behaviour of the system.
56
Thursday, 9 July 2009
/203
Scalability through Modularity
Cellular functions arise from orchestrated interactions between motifs consisting of many molecular interacting species.
A P System model is a set of rules representing molecular interactions motifs that appear in many cellular systems.
57
Thursday, 9 July 2009
/203
Basic P System Modules Used
58
Thursday, 9 July 2009
/203
Modularity in Gene Regulatory Networks
Cis-regulatory modules are nonrandom clusters of target binding sites for transcription factors regulating the same gene or operon.
A P system module is a set of rewriting rules containing variables that can be instantiated with specific objects, stochastic constants and membrane labels.
E. Davidson (2006) The Regulatory Genome, Gene Regulation Networks in Development and Evolution, Elsevier
59
Thursday, 9 July 2009
/203
Modularity in Gene Regulatory Networks
Cis-regulatory modules are nonrandom clusters of target binding sites for transcription factors regulating the same gene or operon.
A P system module is a set of rewriting rules containing variables that can be instantiated with specific objects, stochastic constants and membrane labels.
LuxRAHL
CI
E. Davidson (2006) The Regulatory Genome, Gene Regulation Networks in Development and Evolution, Elsevier
59
Thursday, 9 July 2009
/203
Modularity in Gene Regulatory Networks
Cis-regulatory modules are nonrandom clusters of target binding sites for transcription factors regulating the same gene or operon.
A P system module is a set of rewriting rules containing variables that can be instantiated with specific objects, stochastic constants and membrane labels.
LuxRAHL
CI
E. Davidson (2006) The Regulatory Genome, Gene Regulation Networks in Development and Evolution, Elsevier
59
Thursday, 9 July 2009
/203
Representing transcriptional fusions
Objects Variables can be instantiated with the name of specific genes to represent a construct where the gene is fused to the promoter or cluster of TF binding sites specified by the module.
60
Thursday, 9 July 2009
/203
Representing transcriptional fusions
Objects Variables can be instantiated with the name of specific genes to represent a construct where the gene is fused to the promoter or cluster of TF binding sites specified by the module.
60
Thursday, 9 July 2009
/203
Representing transcriptional fusions
Objects Variables can be instantiated with the name of specific genes to represent a construct where the gene is fused to the promoter or cluster of TF binding sites specified by the module.
60
Thursday, 9 July 2009
/203
Representing transcriptional fusions
Objects Variables can be instantiated with the name of specific genes to represent a construct where the gene is fused to the promoter or cluster of TF binding sites specified by the module.
60
Thursday, 9 July 2009
/203
Representing transcriptional fusions
Objects Variables can be instantiated with the name of specific genes to represent a construct where the gene is fused to the promoter or cluster of TF binding sites specified by the module.
60
Thursday, 9 July 2009
/203
Representing transcriptional fusions
Objects Variables can be instantiated with the name of specific genes to represent a construct where the gene is fused to the promoter or cluster of TF binding sites specified by the module.
60
Thursday, 9 July 2009
/203
Representing transcriptional fusions
Objects Variables can be instantiated with the name of specific genes to represent a construct where the gene is fused to the promoter or cluster of TF binding sites specified by the module.
60
Thursday, 9 July 2009
/203
Representing Directed EvolutionVariables for stochastic constants can be instantiated
with specific values in order to represent directed evolution.
61
Thursday, 9 July 2009
/203
Representing Directed EvolutionVariables for stochastic constants can be instantiated
with specific values in order to represent directed evolution.
61
Thursday, 9 July 2009
/203
Representing Directed EvolutionVariables for stochastic constants can be instantiated
with specific values in order to represent directed evolution.
A
61
Thursday, 9 July 2009
/203
Representing Directed EvolutionVariables for stochastic constants can be instantiated
with specific values in order to represent directed evolution.
A
61
Thursday, 9 July 2009
/203
Representing synthetic transcriptional networks
The genes used to instantiate variables in our modules can codify other TFs that interact with other modules or promoters producing a synthetic gene regulatory network.
62
Thursday, 9 July 2009
/203
Representing synthetic transcriptional networks
The genes used to instantiate variables in our modules can codify other TFs that interact with other modules or promoters producing a synthetic gene regulatory network.
62
Thursday, 9 July 2009
/203
Stochastic P Systems Gillespie Algorithm (SSA) generates trajectories of a stochastic
system consisting of modified for multiple compartments/volumes:
1) A stochastic constant is associated with each rule.2) A propensity is computed for each rule by multiplying the
stochastic constant by the number of distinct possible combinations of the elements on the left hand side of the rule.
3) The rule to apply j0 and the waiting time τ for its application are computed by generating two random numbers r1,r2 ~ U(0,1) and using the formulas:
F. J. Romero-Campero, J. Twycross, M. Camara, M. Bennett, M. Gheorghe, and N. Krasnogor. Modular assembly of cell systems biology models using p systems. International Journal of Foundations of Computer Science, 2009
63
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
Local Gillespie
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
Local Gillespie
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
( 2, τ2, r02)
Local Gillespie
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
( 2, τ2, r02)
( 3, τ3, r03)
Local Gillespie
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
( 2, τ2, r02)
( 3, τ3, r03)
Sort Compartments τ2 < τ1 < τ3
Local Gillespie
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
( 2, τ2, r02)
( 3, τ3, r03)
( 2, τ2, r02)
( 1, τ1, r01)
( 3, τ3, r03)
Sort Compartments τ2 < τ1 < τ3
Local Gillespie
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
( 2, τ2, r02)
( 3, τ3, r03)
( 2, τ2, r02)
( 1, τ1, r01)
( 3, τ3, r03)
Sort Compartments τ2 < τ1 < τ3
Local Gillespie
‘
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
( 2, τ2, r02)
( 3, τ3, r03)
( 2, τ2, r02)
( 1, τ1, r01)
( 3, τ3, r03)
Sort Compartments τ2 < τ1 < τ3
Local Gillespie
( 1, τ1-τ2, r01)
( 3, τ3-τ2, r03)
Update Waiting Times
‘
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
( 2, τ2, r02)
( 3, τ3, r03)
( 2, τ2, r02)
( 1, τ1, r01)
( 3, τ3, r03)
Sort Compartments τ2 < τ1 < τ3
Local Gillespie
( 1, τ1-τ2, r01)
( 3, τ3-τ2, r03)
Update Waiting Times
( 2, τ2’, r02)
‘
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
( 2, τ2, r02)
( 3, τ3, r03)
( 2, τ2, r02)
( 1, τ1, r01)
( 3, τ3, r03)
Sort Compartments τ2 < τ1 < τ3
Local Gillespie
( 1, τ1-τ2, r01)
( 3, τ3-τ2, r03)
Update Waiting Times
( 2, τ2’, r02)
Insert new triplet τ1-τ2 <τ2’ < τ3-τ2
‘
64
Thursday, 9 July 2009
/203
Multicompartmental Gillespie Algorithm
1
2
3 r11,…,r1
n1
M1
r21,…,r2
n2
M2
r31,…,r3
n3
M3
( 1, τ1, r01)
( 2, τ2, r02)
( 3, τ3, r03)
( 2, τ2, r02)
( 1, τ1, r01)
( 3, τ3, r03)
Sort Compartments τ2 < τ1 < τ3
Local Gillespie
( 1, τ1-τ2, r01)
( 3, τ3-τ2, r03)
Update Waiting Times
( 2, τ2’, r02)( 1, τ1-τ2, r0
1)
( 2, τ2’, r02)
( 3, τ3-τ2, r03)
Insert new triplet τ1-τ2 <τ2’ < τ3-τ2
‘
64
Thursday, 9 July 2009
/203
Using P systems modules one can model a large variety of commonly occurring BRN:
Gene Regulatory Networks Signaling Networks Metabolic Networks
This can be done in an incremental way.
F. J. Romero-Campero, J. Twycross, M. Camara, M. Bennett, M. Gheorghe, and N. Krasnogor. Modular assembly of cell systems biology models using p systems. International Journal of Foundations of Computer Science, 2009
65
Thursday, 9 July 2009
/203
InfoBiotics Pipeline
66
Thursday, 9 July 2009
/203
SBML from CellDesigner
67
Thursday, 9 July 2009
/203
Runs simulations and extract data
68
Thursday, 9 July 2009
/203
Plot Timeseries
69
Thursday, 9 July 2009
/203
in time and space
70
Thursday, 9 July 2009
/203
Synthetic Biology Examples
71
Thursday, 9 July 2009
/203
Multi-component negative-feedback oscillator
Oscillations caused by time-delayed negative-feedback:
Negative-feedback: gene-product that represses it's geneTime-delay: mRNA export, translation and repressor import
Novak & Tyson: Design Principles of Biochemical Oscillators. Nat. Rev. Mol. Cell. Biol. 9: 981-991 (2008)
72
Thursday, 9 July 2009
/203
Multi-component negative-feedback oscillator
Mathematical model− Xc = [mRNA in cytosol]− Yc = [protein in cytosol]− Xn = [mRNA in nucleus]− Yn = [protein in nucleus]− E = [total protease]− p = “integer indicating
whether Y binds to DNA as a monomer, trimer, or so on”
Executable Biology makes this more obvious:
we can vary the value of p and the sequence of binding...
73
Thursday, 9 July 2009
/203
Single protein represses genep = 1
74
Thursday, 9 July 2009
/203
When repression is weak(dissociation rate = 10)
No obvious oscillatory behaviour in single simulation75
Thursday, 9 July 2009
/203
When repression is weak(dissociation rate = 10)
Mean of 100 runs shows convergence to steady state76
Thursday, 9 July 2009
/203
When repression is strong(dissociation rate = 0.1)
Oscillations evident in single simulation77
Thursday, 9 July 2009
/203
When repression is strong(dissociation rate = 0.1)
Averging 100 runs dampens oscillations due to different phases but observable. Protein levels steady.
78
Thursday, 9 July 2009
/203
Repressor binding sequence When p=2 there are two possible scenarios:
– First protein binds to second protein weakly then protein-dimer binds to gene strongly
– First protein binds to gene weakly then second protein binds to protein-gene dimer strongly
In the following only the model structure is changed, not the parameters
First dissociation rate = 10 Second dissociation rate = 0.1
79
Thursday, 9 July 2009
/203
1. Protein represses as dimer
80
Thursday, 9 July 2009
/203
1. Protein represses as dimer
target
mRNA levels oscillate ready but protein accumulates in the cytosol
81
Thursday, 9 July 2009
/203
2. Proteins repress cooperatively
82
Thursday, 9 July 2009
/203
2. Proteins repress cooperatively
Oscillations are steady and protein levels are controlled
target
83
Thursday, 9 July 2009
/203
Two different bacterial strains carrying specific synthetic gene regulatory networks are used.
The first strain produces a diffusible signal AHL.
The second strain possesses a synthetic gene regulatory network which produces a pulse of GFP after AHL sensing.
These two bacterial strains and their respective synthetic networks are modelled as a combination of modules.
An example: Ron Weiss' Pulse Generator
S. Basu, R. Mehreja, et al. (2004) Spatiotemporal control of gene expression with pulse generating networks, PNAS, 101, 6355-6360
84
Thursday, 9 July 2009
/203
Sending Cells
85
Thursday, 9 July 2009
/203
Sending Cells
85
Thursday, 9 July 2009
/203
Sending Cells
Pconst
85
Thursday, 9 July 2009
/203
Sending Cells
Pconst luxI
85
Thursday, 9 July 2009
/203
Sending Cells
Pconst
Pconst({X = luxI },…)
luxI
85
Thursday, 9 July 2009
/203
Sending Cells
Pconst
LuxI AHL
AHL
Pconst({X = luxI },…)
PostTransc({X=LuxI},{c1=3.2,…})
Diff({X=AHL},{c=0.1})
luxI
85
Thursday, 9 July 2009
/203
Pulse Generating Cells
86
Thursday, 9 July 2009
/203
Pulse Generating Cells
86
Thursday, 9 July 2009
/203
luxRPconst
LuxR
Pulse Generating Cells
86
Thursday, 9 July 2009
/203
luxRPconst
LuxRAHL
AHL
Pulse Generating Cells
86
Thursday, 9 July 2009
/203
luxRPconst gfp
PluxOR1
LuxR GFPAHL
AHL
Pulse Generating Cells
86
Thursday, 9 July 2009
/203
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
Pulse Generating Cells
86
Thursday, 9 July 2009
/203
luxRPconst
cIPlux
gfpPluxOR1
LuxR
CI
GFPAHL
AHL
Pulse Generating Cells
86
Thursday, 9 July 2009
/203
luxRPconst
cIPlux
gfpPluxOR1
LuxR
CI
GFPAHL
AHL
Pconst({X=luxR},…)
PluxOR1({X=gfp},…)
Plux({X=cI},…)
…
…
Diff({X=AHL},…)
Pulse Generating Cells
86
Thursday, 9 July 2009
/203
Spatial Distribution of Senders and Pulse Generators
luxIPconst
LuxI AHL
AHL
87
luxRPconst
cIPlux
gfpPluxOR1
LuxR
CI
GFPAHL
Thursday, 9 July 2009
/203
Spatial Distribution of Senders and Pulse Generators
luxIPconst
LuxI AHL
AHL
AHL
87
luxRPconst
cIPlux
gfpPluxOR1
LuxR
CI
GFPAHL
AHL
Thursday, 9 July 2009
/203
Wave propagation simulation I
SIMULATION I
88
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
CI
89
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
luxIPlux
LuxI
AHL
CI
89
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
luxIPlux
LuxI
AHL
Pconst({X=luxR},…)
CI
89
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
luxIPlux
LuxI
AHL
Pconst({X=luxR},…)
PluxOR1({X=gfp},…)
CI
89
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
luxIPlux
LuxI
AHL
Pconst({X=luxR},…)
PluxOR1({X=gfp},…)
Plux({X=cI},…)
CI
89
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
luxIPlux
LuxI
AHL
Pconst({X=luxR},…)
PluxOR1({X=gfp},…)
Plux({X=cI},…)
…
CI
89
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
luxIPlux
LuxI
AHL
Pconst({X=luxR},…)
PluxOR1({X=gfp},…)
Plux({X=cI},…)
…
…CI
89
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
luxIPlux
LuxI
AHL
Pconst({X=luxR},…)
PluxOR1({X=gfp},…)
Plux({X=cI},…)
…
…
Diff({X=AHL},…)CI
89
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
gfpPluxOR1
LuxR GFPAHL
AHL
luxIPlux
LuxI
AHL
Pconst({X=luxR},…)
PluxOR1({X=gfp},…)
Plux({X=cI},…)
…
…
Diff({X=AHL},…)Plux({X=luxI},…)
CI
89
Thursday, 9 July 2009
/203 36
Wave propagation simulation II
90
SIMULATION II
Thursday, 9 July 2009
/203
Spatial Distribution of Pulse Generators and Seed
luxRPconst
cIPlux
gfpPluxOR1
LuxR
CI
GFPAHL
AHL
luxI
LuxI
AHL
Plux CI
91
Thursday, 9 July 2009
/203 38
Wave propagation with Four Droplets of Signal
92
SIMULATION III
Thursday, 9 July 2009
/203
Pulse Generating Cells With Relay
luxRPconst
cIPlux
PluxOR1
LuxR
CI
AHL
AHL
luxIPlux
LuxI
AHL
PulseGenerator(X ) =
{ Pconst({X=luxR},…) ,
PluxOR1({X},…) ,
Plux({X=cI},…) ,
…
Diff({X=AHL},…) ,
Plux({X=luxI},…) }
93
Thursday, 9 July 2009
/203
Combining Complex Modules
luxRPconst
Plac
lacIPluxOR1
LuxR
LacI
AHL
AHL
Inverter(X ) =
{ Pconst({X=luxR},…) ,
PluxOR1({X=lacI},…) ,
Plac({X},…) ,
…
Diff({X=AHL},…)}
94
Thursday, 9 July 2009
/203
Combining Complex Modules
luxRPconst
Plac
lacIPluxOR1
LuxR
LacI
AHL
AHL
Inverter(X ) =
{ Pconst({X=luxR},…) ,
PluxOR1({X=lacI},…) ,
Plac({X},…) ,
…
Diff({X=AHL},…)}
PulseGenerator({X=lacI})Inverter({X=gfP})
94
Thursday, 9 July 2009
/203 41
Inversion Through a Propagating Wave
SIMULATION IV
95
Thursday, 9 July 2009
/203
Outline•Brief Introduction to Computational Modeling
•Modeling for Top Down SB•Executable Biology•A pinch of Model Checking
•Modeling for the Bottom Up SB•Dissipative Particle Dynamics
•Automated Model Synthesis and Optimisation
•Conclusions96
Thursday, 9 July 2009
/203
Probabilistic Model Checking A precise computational/mathematical model allows
us to perform formal verification techniques: Probabilistic model checking.
Properties are expressed formally using temporal logic and analysed.
The fundamental components of the PRISM language are modules, variables and commands.• A model is composed of a number of modules which can
interact with each other.• A module contains a number of local variables and commands.
97
Thursday, 9 July 2009
/203
P Systems and PRISMP System Component PRISM Component
Membrane ModuleMultisets of Objects Local Variables
Rewriting rules Commands
Rewards/Costs are associated with states and transitions representing the number of objects and the application of rules.
Some Properties:
Expected Number of objects over time: R = ? [ I = T ] Expected Number of rule application over time: R = ? [ C <= T ] Expected Time to reach a state: R = ? [ F molec_1 = K ] Transient properties: P = ? [ true U[t_1 t_2] molec_1 >= K_1 ] Steady State/Long run properties: S = ? [ molec_1 >= K_1]
98
PRISM is used as an example. Other model checkers are more appropriate for larger systems
Thursday, 9 July 2009
/203
Post-transcriptional Regulation
Post-transcriptionalregulation
[ gene ]b [ gene + rna ]b ctrc
[ rna ]b [ ]b c2=0.3465
[ rna ]b [ rna + Protein ]b ctrl
[ Protein ]b [ ]b c4=0.3465
ctrc*ctrl = 1.13
10 proteins in steady State
99
Thursday, 9 July 2009
/203
Post-transcriptional RegulationR = ? [ I = 240 ]
P = ? [ true U[240,240] Proteins = N ]
100
Thursday, 9 July 2009
/203
Positive Regulation
[ TF + gene ] b [ TF.gene ] b con
[ TF.gene ] b [ TF + gene ] b coff
[ gene ]b [ gene + rna ]b ctrc
[ rna ]b [ ]b c2
[ rna ]b [ rna + Protein ]b ctrl
[ Protein ]b [ ]b c4
101
Thursday, 9 July 2009
/203
Positive RegulationR = ? [ C <= 100 ] R = ? [ C <= 100 ]
P = ? [ true U[60,60] Proteins = N ]
102
Thursday, 9 July 2009
/203
Model Checking on the Pulse Generator
The simulation of the Pulse Generator show some interesting properties that were subsequently analysed using model checking.
Due to the complexity of the system (state space explosion) we perform approximate model checking with a precision of 0.01 and a confidence of 0.001 which needed to run 100000 simulations.
103
Thursday, 9 July 2009
/203
Model Checking on the Pulse Generator
The simulations show that although the number of signals reaches eventually the same level in all the cells in the lattice those cells that are far from the sending cells produce fewer number of GFP molecules.
The difference between cells close to and far from the sending cells is the rate of increase of the signal AHL.
We study the effect of the rate of increase of the signal AHL in the number of GFP produced.
S. Basu, R. Mehreja, et al. Spatiotemporal control of gene expression with pulse generating networks, PNAS, 101, 6355-6360
104
Thursday, 9 July 2009
/203
We studied the expected number of GFP molecules produced over time for different increase rates of AHL.
R = ? [ I = 60 ]
rewards molecule = 1 : proteinGFP;endrewards
The system is expected to produce longer pulses with lower amplitudes for slow increase rates of AHL signals.
105
Thursday, 9 July 2009
/203
In order to get a clearer idea, the probability distribution of the number of GFP molecules at 60 minutes was computed.
P = ? [ true U[60,60] ((proteinGFP > N) & (proteinGFP <= (N + 10))) ]
Note that for slow increase rates of AHL the probability of having NO GFP molecules at all is high.
106
Thursday, 9 July 2009
/203
Finally, assuming that for a cell to be fluorescence it needs to have a given number of GFP for an appreciable period of time we studied the expected amount of time a cell have more than 50 GFP molecules during the first 60 minutes after the signals arrive to the cell.
R = ? [ C <= 60 ]
rewards true : proteinGFP;endrewards
107
Thursday, 9 July 2009
/203
Outline•Brief Introduction to Computational Modeling
•Modeling for Top Down SB•Executable Biology•A pinch of Model Checking
•Modeling for the Bottom Up SB•Dissipative Particle Dynamics
•Automated Model Synthesis and Optimisation
•Conclusions108
Thursday, 9 July 2009
/203
A (Proto)Cell as an Information Processing Device
LeDuc et al. Towards an in vivo biologically inspired nanofactory. Nature (2007)
109
Thursday, 9 July 2009
/203
Towards a synthetic cell from the bottom up
Biocompatible vesicles as long-circulating carriers Polymer self-assembly into higher-order structures Cell-mimics with hydrophobic ‘cell-wall’ and glycosylated
surfaces Potential for cross-talk with biological cells
Pasparakis, G. Angew Chem Int Ed. 2008 47 (26), 4847-4850
110
Thursday, 9 July 2009
/203
Vesicle Biorecognition
Pasparakis, G. et al, Angew Chem Int Ed. 2008 47 (26), 4847-4850
111
Thursday, 9 July 2009
/203
‘Talking’ to cell-vesicle aggregates
Pasparakis, G. Angew Chem Int Ed. 2008 47 (26), 4847-4850
112
Thursday, 9 July 2009
/203
Outline•Brief Introduction to Computational Modeling
•Modeling for Top Down SB•Executable Biology•A pinch of Model Checking
•Modeling for the Bottom Up SB•Dissipative Particle Dynamics
•Automated Model Synthesis and Optimisation
•Conclusions113
Thursday, 9 July 2009
/203
Dissipative Particle Dynamics Simulate movement of particles which represent several
atoms / molecules Calculate forces acting on particles, integrate equations of
motion Used extensively for investigating the self-assembly of lipid
membrane structures at the mesoscale Typical simulations contain ~105-106 particles, for ~105-106 time
steps Particles interact with each other within a finite radius much
smaller than the simulation space, algorithmic optimisations of force calculations are possible
114
Thursday, 9 July 2009
/203
Dissipative Particle Dynamics First introduced by Hoogerbrugge and Koelmann in 1992. Statistical mechanics of the model derived by espanol and warren in
1995. A coarse graining approach is used so that one simulation particle
represents a number of real molecules of a given type. Since the timescale at which interactions occur is longer than in MD,
fewer time-steps are required to simulation the same period of real time. The short force cut-off radius enables optimisation of the force calculation
code to be performed.
H HO
H HO
H HO
W
115
Thursday, 9 July 2009
/203
Dissipative Particle Dynamics
W
W
i
j
P
P
i
j
Conservative Force
Dissipative Force
Random Force
116
Thursday, 9 July 2009
/203
Dissipative Particle Dynamics Polymers A number of simulation beads are tied together to
represent the original molecule. Two new forces are introduced between polymer
particles, a Hookean spring force and a bond angle force.
117
Thursday, 9 July 2009
/203
Liposome Formation in DPD
118
Thursday, 9 July 2009
/203119
Thursday, 9 July 2009
/203
Case Study One: Vesicle Diffusion
Polar heads
Non polar tails
Pores
J. Smaldon, J. Blake, D. Lancet, and N. Krasnogor. A multi-scaled approach to artificial life simulation with p systems and dissipative particle dynamics. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO-2008), ACM Publisher, 2008.
120
Thursday, 9 July 2009
/203
Case Study One: Vesicle Diffusion
The regions were formed by allowing vesicles to self-assemble from phospholipids in the presence of pore inclusions
Pores are simple channels with an exterior mimicking the hydrophobic/hydrophilic profile of the bilayer
121
Thursday, 9 July 2009
/203122
Thursday, 9 July 2009
/203
Case Study One: Vesicle DiffusionTagged solvent particles were placed within the liposome inner volume, the change in concentration due to diffusion of solvent through the membrane pores was measures
123
Thursday, 9 July 2009
/203
Case Study Two: Liposome Logic
The behaviour of some prokaryotic RNA transcription motifs matches that of boolean logic gates[1]
DPD was extended with mesoscale collision based reactions.
transcriptional logic gates were simulated in bulk solvent and within a liposome core volume.
124
Thursday, 9 July 2009
/203125
Thursday, 9 July 2009
/203
Case Study Two: Liposome LogicOR gate results for different inputs: (¬X,¬Y) (¬X,Y) (X,¬Y) (X,Y)
J. Smaldon, N. Krasnogor, M. Gheorghe, and A. Cameron. Liposome logic. In Proceedings of the 2009 Genetic and Evolutionary Computation Conference (GECCO 2009), 2009
126
Thursday, 9 July 2009
/203
Parallel DPD In order to simulate large systems, the
DPD method can be parallelised using spatial decomposition techniques
Each processor is assigned a subvolume of the 3D simulation space
Communication required when particles move between subvolumes
Communication performed using MPI 10x-15x faster than the single processor
version of DPD (Jupiter implementation)
127
Thursday, 9 July 2009
/203
Compute Unified Device Architecture (CUDA) DPD
Algorithms can be written in C code for execution on an Nvidia G8X/GT200 GPU.
Modern CPUs have 2-4 cores, modern GPUs have hundreds. Highly parallel SIMD computing.
Several orders of magnitude calculation speed increase for MD.
The DPD algorithm was expressed in a manner enabling parallel execution with CUDA.
128
Thursday, 9 July 2009
/203
CUDA DPD Implementation
CUDA DPD is ~27 times faster than single processor
129
Thursday, 9 July 2009
/203
Outline•Brief Introduction to Computational Modeling
•Modeling for Top Down SB•Executable Biology•A pinch of Model Checking
•Modeling for the Bottom Up SB•Dissipative Particle Dynamics
•Automated Model Synthesis and Optimisation
•Conclusions130
Thursday, 9 July 2009
/203
Automated Model Synthesis and Optimisation
Modeling is an intrinsically difficult process
It involves “feature selection” and disambiguation
Model Synthesis requires design the topology or structure of the system in
terms of molecular interactions estimate the kinetic parameters associated with
each molecular interaction
All the above iterated131
Thursday, 9 July 2009
/203
Once a model has been prototyped, whether derived from existing literature or “ab initio” ➡ Use some optimisation method to fine tune parameters/model structure
adopts an incremental methodology, namely starting from very simple P system modules (BioBricks) specifying basic molecular interactions, more complicated modules are produced to model more complex molecular systems.
132
Thursday, 9 July 2009
/203
Large Literature on Model Synthesis• Mason et al. use a random Local Search (LS) as the mutation to
evolve electronic networks with desired dynamics
• Chickarmane et al. use a standard GA to optimize the kinetic parameters of a population of ODE-based reaction networks having the desired topology.
• Spieth et al. propose a memetic algorithm to find gene regulatory networks from experimental DNA microarray data where the network structure is optimized with a GA and the parameters are optimized with an Evolution Strategy (ES).
• Etc
133
Thursday, 9 July 2009
/203
Evolutionary Algorithms for Automated Model Synthesis and Optimisation
EA are potentially very useful for AMSO There’s a substantial amount of work on:
using GP-like systems to evolve executable structures
using EAs for continuous/discrete optimisation
An EA population represents alternative models (could lead to different experimental setups)
EAs have the potential to capture, rather than avoid, evolvability of models
134
Thursday, 9 July 2009
/203
• The main idea is to use a nested evolutionary algorithm where the first layer evolves model structures while the inner layer acts as a local search for the parameters of the model. • It uses stochastic P systems as a computational, modular and discrete-stochastic modelling framework.
• It adopts an incremental methodology, namely starting from very simple P system modules specifying basic molecular interactions, more complicated modules are produced to model more complex molecular systems.
•Successfully validated evolved models can then be added to the models library
Evolving Executable Biology Models
135
Thursday, 9 July 2009
/203
Nested EA for Model Synthesis
F. Romero-Campero, H.Cao, M. Camara, and N. Krasnogor. Structure and parameter estimation for cell systems biology models. Proceedings of the Genetic and Evolutionary Computation Conference (GECCO-2008), pages 331-338. ACM Publisher, 2008.
Best Paper award at the Bioinformatics track.
136
Thursday, 9 July 2009
/203
Fitness Evaluation
137
Thursday, 9 July 2009
/203
Representation
138
Thursday, 9 July 2009
/203139
Thursday, 9 July 2009
/203
Structural Operators
140
Thursday, 9 July 2009
/203
Structural Operators
141
Thursday, 9 July 2009
/203
Parameter Optimisation
142
Thursday, 9 July 2009
/203
Parameter Optimisation
143
Thursday, 9 July 2009
/203
Simple Illustrative Case studies
• Case 1: molecular complexation
• Case 2: enzymatic reaction
• Case 3: autoregulation in transcriptional networks
•Case 3.1. negative autoregulation•Case 3.2. positive autoregulation
144
Thursday, 9 July 2009
/203
Case 1: Molecular ComplexationTarget P System Model Structure
Target P System Model Parameters
c1
c2
145
Thursday, 9 July 2009
/203
Case 1: Experimental Results
50/50 runs found the same P system model structure as the target one
Target and Evolved Time Series 146
Thursday, 9 July 2009
/203
Case 1: Experimental Results
50/50 runs found the same P system model structure as the target one
Best Model Parameters
Target Model Parameters
Target and Evolved Time Series 147
Thursday, 9 July 2009
/203
Case 2: Enzymatic ReactionTarget P System Model Structure
Target P System Model Parameters
c1
c2
c3
148
Thursday, 9 July 2009
/203
Case 2: Experimental ResultsTarget and Evolved Time Series
149
Thursday, 9 July 2009
/203
Case 2: Experimental Results
Best Model Parameters
Target Model Parameters Target and Evolved Time Series
Two Approaches Using Basic Modules
Adding the Newly Found in Case 1
Number of runs the target P system model was
found
12/50 39/50
Mean RMSE 3.8 2.85150
Thursday, 9 July 2009
/203
Case 3.1: Negative AutoregulationTarget P System Model Structure Target P System Model Parameters
c1
c2
c3
c4
c5
c6
151
Thursday, 9 July 2009
/203
Case 3.1: Experimental Results
Best Model in Design 1
Design Design 1 Design 2
Model
Number of runs 46/50 4/50
Mean +/- STD 4.11+/- 11.73 4.63 +/- 2.91
Best Model in Design 2
152
Thursday, 9 July 2009
/203
Case 3.2: Positive AutoregulationTarget P System Model Structure Target P System Model Parameters
c1
c2
c3
c4
c5
c6
c7
153
Thursday, 9 July 2009
/203
Case 3.2: Experimental ResultsDesign Design 1 Design 2
Model
Number of runs 30/50 20/50
Mean +/- STD 16.36 +/- 3.03 20.14 +/- 5.13
Best Model in Design 1
154
Thursday, 9 July 2009
/203
Revisiting the Fitness Function• Multiple time-series per target
• Different time series have very different profiles, e.g., response time or maxima occur at different times/places
• Transient states (sometimes) as important as steady states
•RMSE will mislead search
H. Cao, F. Romero-Campero, M.Camara, N.Krasnogor. Analysis of Alternative Fitness Methods for the Evolutionary Synthesis of Cell Systems Biology Models. Submitted (2009)
155
Thursday, 9 July 2009
/203
Four Fitness FunctionsN time seriesM time points in each time series Simulated data Target data
F1
F2
F2 normalises whithin a time series between [0,1]
156
Thursday, 9 July 2009
/203
Four Fitness FunctionsN time seriesM time points in each time series Simulated data Target data Randomly generated normalised
vector with:
F3
F3, unlike F1, does not assume an equal weighting of all the errors. It produces a randomised average of weights.
157
Thursday, 9 July 2009
/203
Four Fitness FunctionsN time seriesM time points in each time series Simulated data Target data
F4
F4 simply multiplies errors hence even small ones within a set with multiple orders of maginitudes can still have an effect. Indeed, this might lead to numerical instabilities due to nonlinear effects.
158
Thursday, 9 July 2009
/203
Further Case Studies
159
Thursday, 9 July 2009
/203160
Thursday, 9 July 2009
/203161
Thursday, 9 July 2009
/203162
Thursday, 9 July 2009
/203163
Thursday, 9 July 2009
/203164
Thursday, 9 July 2009
/203165
Thursday, 9 July 2009
/203166
Thursday, 9 July 2009
/203
Results Study Case 1
167
Thursday, 9 July 2009
/203
Results Study Case 2
168
Thursday, 9 July 2009
/203
Results Study Case 3
169
Thursday, 9 July 2009
/203
Alternative, biologically valid, modules
Target model
170
Thursday, 9 July 2009
/203171
Thursday, 9 July 2009
/203
Results Study Case 4
172
Thursday, 9 July 2009
/203
Comparisons of the constants between the best fitness models obtained by F1, F4 and the target model for Test Case 4 (Values different from the target are in bold and underlined)
173
Thursday, 9 July 2009
/203174
Thursday, 9 July 2009
/203175
Thursday, 9 July 2009
/203176
Thursday, 9 July 2009
/203177
Thursday, 9 July 2009
/203178
Thursday, 9 July 2009
/203
Target
179
Thursday, 9 July 2009
/203
Target
180
Thursday, 9 July 2009
/203
The evolving curve of average fitness by four fitness methods (F1 ∼ F4) in 20 runs for Test Case 4.
181
Thursday, 9 July 2009
/203
The evolving curve of average model diversity by four fitness methods (F1 ∼ F4) in 20 runs for Test Case 4
182
Thursday, 9 July 2009
/203183
Thursday, 9 July 2009
/203
Multi-Objective Optimisation in Morphogenesis
The following slides are based on
Rui Dilão, Daniele Muraro, Miguel Nicolau, Marc Schoenauer. Validation of a morphogenesis model of Drosophila early development by a multi-objective evolutionary optimization algorithm. Proc. 7th European Conference on Evolutionary Computation, ML and Data Mining in BioInformatics (EvoBIO'09), April 2009.
184
Thursday, 9 July 2009
/203
• The authors investigate the use of MOEA for modeling morphogenesis
• The model organism is drosophila embrios Rui Dilão, Daniele Muraro, Miguel Nicolau, Marc Schoenauer. Validation of a morphogenesis model of Drosophila early development by a multi-objective evolutionary optimization algorithm. Proc. 7th European Conference on Evolutionary Computation, ML and Data Mining in BioInformatics (EvoBIO'09), April 2009
185
Thursday, 9 July 2009
/203
Initial Conditions
PDE model
186
Thursday, 9 July 2009
/203
Target
By Evolving:abcd, acad, Dbcd/Dcad, r, mRNA distributions and t
However: Goal is not perfect fit but rather robust fit
187
Thursday, 9 July 2009
/203
• The authors use both Single and Multi-objective optimisation• CMA-ES is at the core of both• CMA-ES is a (µ,λ)-ES that employs multivariate Gaussian
distributions • Uses “cumulative path” for co-variantly adapting this
distribution• For the MO case it uses the global pareto dominance based
selection
Bi-Objective Function
Objective Function
188
Thursday, 9 July 2009
/203189
Thursday, 9 July 2009
/203190
Thursday, 9 July 2009
/203191
Thursday, 9 July 2009
/203
Outline•Brief Introduction to Computational Modeling
•Modeling for Top Down SB•Executable Biology•A pinch of Model Checking
•Modeling for the Bottom Up SB•Dissipative Particle Dynamics
•Automated Model Synthesis and Optimisation
•Conclusions192
Thursday, 9 July 2009
/203
Summary & Conclusions This talk has focused on an integrative methodology,
InfoBiotics, for Systems & Synthetic Biology Executable Biology/DPD Parameter and Model Structure Discovery Model Checking
Computational models (or executable in Fisher & Henzinger’s jargon) adhere to (a degree) to an operational semantics.
Refer to the excellent review [Fisher & Henzinger, Nature Biotechnology, 2007]
193
Thursday, 9 July 2009
/203
The gap present in mathematical models between the model and its algorithmic implementation disappears in computational models as all of them are algorithms.
A new gap appears between the biology and the modelling technique and this can be solved by a judicious “feature selection”, i.e. the selection of the correct abstractions
Good computational models are more intuitive and analysable
Summary & Conclusions
194
Thursday, 9 July 2009
/203
Computational models can thus be executed (quite a few tools out there, lots still missing)
Quantitative VS qualitative modelling: computational models can be very useful even when not every detail about a system is known.
Missing Parameters/model structures can sometimes be fitted with of-the-shelf optimisation strategies (e.g. COPASI, GAs, etc)
Computational models can be analysed by model checking: thus they can be used for testing hypothesis and expanding experimental data in a principled way
Summary & Conclusions
195
Thursday, 9 July 2009
/203
A nested evolutionary algorithm is proposed to automatically develop and optimise the modular structure and parameters of cellular models based on stochastic P systems. Several case studies with incremental model complexity demonstrate the effectiveness of our algorithm.
The fact that this algorithm produces alternative models for a specific biological signature is very encouraging as it could help biologists to design new experiments to discriminate among competing hypothesis (models).
Comparing results by only using the elementary modules and by adding newly found modules to the library shows the obvious advantage of the incremental methodology with modules. This points out the great potential to automatically design more complex cellular models in the future by using a modular approach.
Summary & Conclusions
196
Thursday, 9 July 2009
/203
Synthetising Synthetic Biology Models is more like evolving general GP programs and less like fitting regresion or inter/extra-polation
We evolve executable structures, distributed programs(!) These are noisy and expensive to execute Like in GP programs, executable biology models might achieve
similar behaviour through different program “structure” Prone to bloat Like in GP, complex relation between diversity and solution
quality However, diverse solutions of similar fit might lead to interesting
experimental routes Co-desig of models and wetware.
Summary & Conclusions
197
Thursday, 9 July 2009
/203
Summary & Conclusions Some really nice tutorials and other sources:
Luca Caderlli’s BraneCalculus & BioAmbients Simulating Biological Systems in the Stochastic π−calculus by Phillips and Cardelli
From Pathway Databases to Network Models by Aguda and Goryachev
Modeling and analysis of biological processes by Brane Calculi and Membrane Systems by Busi and Zandron
D. Gilbert’s website contain several nice papers with related methods and tutorials
198
Thursday, 9 July 2009
/203
Other SourcesF. J. Romero-Campero, J. Twycross, M. Camara, M. Bennett, M. Gheorghe, and N. Krasnogor. Modular assembly of cell systems biology models using p systems. International Journal of Foundations of Computer Science, (to appear), 2009.
F.J. Romero-Camero and N. Krasnogor. An approach to biomodel engineering based on p systems. In Proceedings of Computation In Europe (CIE 2009), 2009.
J. Smaldon, N. Krasnogor, M. Gheorghe, and A. Cameron. Liposome logic. In Proceedings of the 2009 Genetic and Evolutionary Computation Conference (GECCO 2009), 2009
F. Romero-Campero, H.Cao, M. Camara, and N. Krasnogor. Structure and parameter estimation for cell systems biology models. In Maarten Keijzer et.al, editor, Proceedings of the Genetic and Evolutionary Computation Conference (GECCO-2008), pages 331-338. ACM Publisher, 2008. This paper won the Best Paper award at the Bioinformatics track.
J. Smaldon, J. Blake, D. Lancet, and N. Krasnogor. A multi-scaled approach to artificial life simulation with p systems and dissipative particle dynamics. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO-2008). ACM Publisher, 2008.
199
Thursday, 9 July 2009
/203
Other SourcesPăun, Gh. Computing with membranes. Journal of Computer and System Sciences 61 (2000) 108-143
P Systems Web Page http://psystems.disco.unimib.it/ Bianco L. Membrane Models of Biological Systems PhD thesis 2007
Bernardini F, Gheorghe M, Krasnogor N, Terrazas G. Membrane Computing - Current Results and Future Problems. CiE 2005 49-53
Bernardini F, Gheorghe M, Krasnogor N. Quorum sensing P systems. Theoretical Computer Science 371 (2007) 20-33
Miguel Nicolau, Marc Schoenauer. Evolving Specific Network Statistical Properties using a Gene Regulatory Network Model. In ECCS 2008, 5th European Conference on Complex Systems, 2008. Miguel Nicolau, Marc Schoenauer. Evolving Scale-Free Topologies using a Gene Regulatory Network Model. CEC 2008, IEEE Congress on Evolutionary Computation, pp. 3748-3755, IEEE Press, 2008.
200
Thursday, 9 July 2009
/203
A. Ridwan. A parallel implementation of Gillespie's Direct Method. Proc. of the International Conference on Computational Science, p.284-291, Krakow, Poland, June 2004.
G. C. Ewing et al. Akaroa2: Exploiting network computing by distributing stochastic simulation. Proc. of the European Simulation Multiconference, p.175-181, Warsaw, June 1999.
P. Hellekalek. Don't trust parallel Monte Carlo! ACM SIGSIM Simulation Digest, 28(1):82-89, 1998.
M. Schwehm. Parallel stochastic simulation of whole cell models. Proc. of the Second International Conference on Systems Biology, p.333-341, CalTech, C.A., November 2001
Other Sources
201
Thursday, 9 July 2009
/203
Acknowledgements
Jonathan Blake
Hongqing Cao
Francisco Romero-Campero
James Smaldon
Jamie Twycross
Integrated Environment
Machine Learning & Optimisation
Modeling & Model Checking
Dissipative Particle Dynamics
Stochastic Simulations
Members of my team working on SB2
EP/E017215/1
EP/D021847/1
BB/F01855X/1
BB/D019613/1
My colleagues in the Centre for Biomolecular Sciences and the Centre for Plant Integrative Biology at Nottingham
• GECCO 2009 organisers for inviting this tutorial, specially Martin Butz and Gunther Raidl
• You for listening!202
Thursday, 9 July 2009
/203
Any Questions?
•www.infobiotic.org
•www.synbiont.org Become a member and have access to $$$ forengaging in SB research. Contact me if interested
203
Thursday, 9 July 2009