Towards a Rapid Model Prototyping Strategy for Systems & Synthetic Biology

BEACON, Michigan State University, July 2010 /82

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

Towards a Rapid Model Prototyping Strategy for Systems & Synthetic Biology

1


Outline•Conceptual Viewpoint & CAD tools

•Model Specification Framework

•Examples of Modeling for Top Down SB

• In Silico Model Evolution

•Conclusions

2


A (Proto)Cell as an Information Processing Device

LeDuc et al. Towards an in vivo biologically inspired nanofactory. Nature (2007)

3


Functional EntitiesContainer

• A boundary defining self/non-self (symmetry breaking).• Maintain concentration gradients and avoid environmental damage.

Metabolism

• Energy utilisation • Confining raw materials to be processed.• Maintenance of internal structures (autopoiesis).

Information

• Sensing environmental signals / release of signals.• Genetic information

4


•The Cell senses the environment and its own internal states•Makes Plans, Takes Decisions and Acts•Evolution is the master programmer

5

The Cell as an Intelligent (Evolved) Machine

Cell

Internal States

Environmental Inputs

Actions

Amir Mitchell, et al., Adaptive prediction of environmental changes by microorganisms. Nature June 2009.


•The Cell senses the environment and its own internal states•Makes Plans, Takes Decisions and Acts•Evolution is the master programmer

5

The Cell as an Intelligent (Evolved) Machine

Cell

Internal States

Environmental Inputs

Actions

Amir Mitchell, et al., Adaptive prediction of environmental changes by microorganisms. Nature June 2009.

Wikimedia Commons


Network Motifs: Evolution’s Preferred Circuits•Biological networks are complex and vast•To understand their functionality in a scalable way one must choose the correct abstraction

•Moreover, these patterns are organised in non-trivial/non-random hierarchies

•Each network motif carries out a specific information-processing function

6

“Patterns that occur in the real network significantly more often than in randomized networks are called network motifs” Shai S. Shen-Orr et al., Network motifs in the transcriptional regulation

network of Escherichia coli. Nature Genetics 31, 64 - 68 (2002)

Radu Dobrin et al., Aggregation of topological motifs in the Escherichia coli transcriptional regulatory network. BMC Bioinformatics. 2004; 5: 10.


Y positively regulates X

Negative autoregulation

Positive autoregulation

The C1-FFL is a ‘sign-sensitive delay’ element and a persistence detector.The I1-FFL is a pulse generator and response accelerator

U. Alon. Network motifs: theory and experimental approaches. Nature Reviews Genetics (2007) vol. 8 (6) pp. 450-461

Shai S. Shen-Orr et al., Network motifs in the transcriptional regulation network of Escherichia coli. Nature Genetics 31, 64 - 68 (2002)




•Cells (and most biologists) don’t do differential calculus! •P systems are a executable specifications that closely mimic biological reality.•These are programs that explicitly mimic the internal behavior of cell systems.•These programs are executed in a virtual machine that captures the intrinsic stochasticity inherent in biology

8


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)9

A. Regev, E. Shapiro. The π-calculus as an abstraction for biomolecular systems. Modelling in Molecular Biology., pages 1–50. Springer Berlin., 2004.

D. Harel, "A Grand Challenge for Computing: Full Reactive Modeling of a Multi-Cellular Animal", Bulletin of the EATCS , European Association for Theoretical Computer Science, no. 81, 2003, pp. 226-235


•A readable language but still a formal language with unambiguous semantics•Close to graphical depictions of metabolic networks normally used by biologists

Erythrocyte cell’s glycolytic and pentose phospate pathway from [Reddy et al., Comput. Biol. Med., 1996]

Properties of Petri Net Models

10


•A readable language but still a formal language with unambiguous semantics•Close to graphical depictions of metabolic networks normally used by biologists

Erythrocyte cell’s glycolytic and pentose phospate pathway from [Reddy et al., Comput. Biol. Med., 1996]

Properties of Petri Net Models

10


Some Limitations

•Not designed to handle spatial events or spatial processes such as diffusion, mechanical transduction, Brownian motion, collisions, transport, etc.•Stochasticity is loaded onto W, i.e. it is not a property of rules or interactions•Although very elegant computing invariants and other properties can really only be done for small systems.•Although due to composability, to certain extent, you can “divide & conquer”

11


Elementary Concepts in π-calculus

•A program specifies a network of interacting processes•Processes are defined by their potential communication activities •Communication occurs on complementary channels, identified by names•Message content: Channel name

12

Biological reactions are thusabstracted as communications.


π-calculusProcess expressionsP ::= 0 a null process π.P P + P mutually exclusive processes P | P concurrent processes (νx)P new channel x in P ! P create a copy of P

Action prefixes π ::= x? (y) receive y along x x! <y> send y along x τ unobservable (internal) action

syntax

13




syntax a molecule / domain

13





A set of molecules / domains

13






a capability to do something(e.g. phosporilate, cleave, etc)

13






a capability to do something(e.g. phosporilate, cleave, etc)

This process (enzyme)knows how to do action π (e.g. ligate)

13


Some Limitations•Non-mobile/mobile approaches offer a tradeoff between relevance and utility.•Computational problem: each molecule is an instance of a process.•Some abstraction are obscure:

•private channels for compartments is cumbersome, requires multi-step encoding of the formation of complexes, not really a nature encoding of proximity and biochemical complementarity.

•Communication always involves exactly two processes.

14


Biochemical Process Calculi

π-calculus (Stochastic, Causal) BioAmbients (mobile compartments) Brane calculi (membranes, Projective) Beta Binders (dynamic compartments)

increasing emphasison intracellular organisation

based on π-calculus

15


•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

BU/TD Synthetic & Systems Biologist

Mathematician

Tools Suitability and Cost

Computer Scientist/Engineer

16


InfoBiotics Workbench and Dashboard

Spec

ifica

tion

Sim

ulat

ion

Analysis

VerificationO

ptimisation

Managementwww.infobiotics.net

http://www.infobiotics.net

http://www.infobiotics.net






•Conclusions

18





Spec

ifica

tion


Model Specification 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

20

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, 20(3):427-442, 2009

G. Paun. Computing with membranes. Journal of Computer and System Sciences, 61(1):108–143, 2000


Distributed and parallel rewritting systems in compartmentalised hierarchical structures.

Compartments

Objects

Rewriting Rules

• Computational universality and efficiency.

• Modelling Framework

21

What is a P Systems?


P-Systems’ Principal Entities

Molecules (ATP, ADP, OHL, etc)

Letters (objects) from an alphabet

Structured Molecules (DNA, RNA,Proteins, etc)

Strings in the alphabet

Molecules in “bulk” Multisets of objects/strings

Membranes/organelles

Membrane

Biochemical activity RulesBiochemical transport Communication rules

22


Rewriting Rules

used by Multi-volume Gillespie’s algorithm

23


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

24


Compartments / Cells Compartments and regions are explicitly

specified using membrane structures.

25


Colonies / Tissues Colonies and tissues are representing as a

collection of P systems distributed over a lattice.

Objects can travel around the lattice through translocation rules.

v

26


Molecular Interactions Inside Compartments

27


Passive Diffusion of Molecules

28


Specification of Transcriptional Regulatory Networks

29


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 may also have (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.

30


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 modules of rules representing molecular interactions (network) motifs that appear in many cellular systems.

31


Basic P System Modules Used

32

BEACON, Michigan State University, July 2010 /82 21

Characterisation/Encapsulation of Cellular Parts: Gene Promoters

A modeling language for the design of synthetic bacterial colonies.

A module, set of rules describing the molecular interactions involving a cellular part, provides encapsulation and abstraction. Collection or libraries of reusable cellular parts and reusable models.

E. Davidson (2006) The Regulatory Genome, Gene Regulation Networks in Development and Evolution, Elsevier

33





01101110100001010100011110001011101010100011010100


33






33





LuxRAHL

CI


33





LuxRAHL

CI

PluxOR1({X},{c1, c2, c3, c4, c5, c6, c7, c8, c9},{l}) = {

type: promoter

sequence: ACCTGTAGGATCGTACAGGTTTACGCAAGAA ATGGTTTGTATAGTCGAATACCTCTGGCGGTGATA

rules: r1: [ LuxR2 + PluxPR.X ]_l -c1-> [ PluxPR.LuxR2.X ]_l r2: [ PluxPR.LuxR2.X ]_l -c2-> [ LuxR2 + PluxPR.X ]_l ... r5: [ CI2 + PluxPR.X ]_l -c5-> [ PluxPR.CI2.X ]_l r6: [ PluxPR.CI2.X ]_l -c6-> [ CI2 + PluxPR.X ]_l ... r9: [ PluxPR.LuxR2.X ]_l -c9-> [ PluxPR.LuxR2.X + RNAP.X ]_l}


33


Module Variables: Recombinant DNA, Directed Evolution, Chassis selection

34



Recombinant DNA: Objects variables can be instantiated with the name of specific genes.

34




34




PluxOR1({X=tetR})

34




34




PluxOR1({X=GFP})

34




PluxOR1({X=GFP})

34



Directed evolution: Variables for stochastic constants can be instantiated with specific values.


PluxOR1({X=GFP})

34



A



PluxOR1({X=GFP})

34



A



PluxOR1({X=GFP})

PluxOR1({X=GFP},{...,c4=10,...})

34



A



PluxOR1({X=GFP})

PluxOR1({X=GFP},{...,c4=10,...}) Chassis Selection: The variable for the label can be instantiated with the name of a

chassis.

PluxOR1({X=GFP},{...,c4=10,...},{l=DH5α })

34


Characterisation/Encapsulation of Cellular Parts: Riboswitches

A riboswitch is a piece of RNA that folds in different ways depending on the presence of absence of specific molecules regulating translation.

ToppRibo({X},{c1, c2, c3, c4, c5, c6},{l}) = {

type: riboswitch

sequence:GGTGATACCAGCATCGTCTTGATGCCCTTGG CAGCACCCCGCTGCAAGACAACAAGATG rules: r1: [ RNAP.ToppRibo.X ]_l -c1-> [ ToppRibo.X ]_l r2: [ ToppRibo.X ]_l -c2-> [ ]_l r3: [ ToppRibo.X + theop ]_l –c3-> [ ToppRibo*.X ]_l r4: [ ToppRibo*.X ]_l –c4-> [ ToppRibo.X + theop ]_l r5: [ ToppRibo*.X ]_l –c5-> [ ]_l r6: [ ToppRibo*.X ]_l –c6-> [ToppRibo*.X + Rib.X ]_l}

35


Characterisation/Encapsulation of Cellular Parts: Degradation Tags

Degradation tags are amino acid sequences recognised by proteases. Once the corresponding DNA sequence is fused to a gene the half life of the protein is reduced considerably.

degLVA({X},{c1, c2},{l}) = {

type: degradation tag

sequence: CAGCAAACGACGAAAACTACGCTTTAGTAGCT

rules: r1: [ Rib.X.degLVA ]_l -c1-> [ X.degLVA ]_l r2: [ X.degLVA ]_l -c2-> [ ]_l}

36


Higher Order Modules: Building Synthetic Gene Circuits

PluxOR1 geneXToppRibo degLVA

37




3OC6_repressible_sensor({X}) = {

37




3OC6_repressible_sensor({X}) = { PluxOR1({X=ToppRibo.geneX.degLVA},{...},{l=DH5α})

37




3OC6_repressible_sensor({X}) = { PluxOR1({X=ToppRibo.geneX.degLVA},{...},{l=DH5α}) ToppRibo({X=geneX.degLVA},{...},{l=DH5α})

37




3OC6_repressible_sensor({X}) = { PluxOR1({X=ToppRibo.geneX.degLVA},{...},{l=DH5α}) ToppRibo({X=geneX.degLVA},{...},{l=DH5α}) degLVA({X},{...},{l=DH5α})

37




3OC6_repressible_sensor({X}) = { PluxOR1({X=ToppRibo.geneX.degLVA},{...},{l=DH5α}) ToppRibo({X=geneX.degLVA},{...},{l=DH5α}) degLVA({X},{...},{l=DH5α})}

37





X=GFP

37





X=GFP

37





X=GFP

Plux({X=ToppRibo.geneCI.degLVA},{...},{l=DH5α}) ToppRibo({X=geneCI.degLVA},{...},{l=DH5α}) degLVA({CI},{...},{l=DH5α})

PtetR({X=ToppRibo.geneLuxR.degLVA},{...},{l=DH5α}) Weiss_RBS({X=LuxR},{...},{l=DH5α}) Deg({X=LuxR},{...},{l=DH5α})

37


luxIPconst

LuxI AHL

AHL

luxRPconst

cIPlux

gfpPluxOR1

LuxR

CI

GFPAHL

AHLSpecification of Multi-cellular

Systems: LPP systems

38


Infobiotics: An Integrated Frameworkhttp://www.infobiotics.org/infobiotics-workbench/

Synthetic Multi-cellular Systems

Libraries of Modules

P systems LPP systems

Multi Compartmental Stochastic Simulations

based on Gillespie’s algorithm

Spatio-temporal Dynamics Analysis

using Model Checking with PRISM and MC2

Automatic Design of Synthetic Gene

Regulatory Circuits using Evolutionary Algorithms

A compiler based on a BNF grammar

Single Cells

Cellular Parts

Synthetic Circuits

Module Combinations

39


Stochastic P Systems Are Executable Programs

The virtual machine running these programs is a “Gillespie Algorithm (SSA)”. It generates trajectories of a stochastic system:

A stochastic constant is associated with each rule.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.

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

40


•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, N. Krasnogor. Modular Assembly of Cell Systems Biology Models Using P Systems. Internatnional Journal of Foundations of Computer Science, 20(3):427-442, 2009.

41

Degano P., Prandi D., Priami C., and Quaglia P. Beta-binders for biological quantitative experiments.ENTCS, 164:101-117, 2006. Proc. 4th Workshop on Quantitative Aspects of Programming Languages, QAPL 2006.






•Conclusions

42



Spec

ifica

tion



Spec

ifica

tion

Sim

ulat

ion

Analysis


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

44


Pconst

LuxI AHL

AHL

Pconst({X = luxI },…)

PostTransc({X=LuxI},{c1=3.2,…})

Diff({X=AHL},{c=0.1})

luxI

luxRPconst

cIPlux

gfpPluxOR

LuxRGFPAHL

AHL

Pconst({X=luxR},…)

PluxOR1({X=gfp},…)

Plux({X=cI},…) …

An example: Ron Weiss' Pulse Generator

Signal Sender Pulse Generator

45

CI


Spatial Distribution of Senders and Pulse Generators

luxIPconst

LuxI AHL

AHL

luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

46

CI


Pulse propagation - simulation I

47

Simulation I

http://localhost/~nxk/MOVIES/pulseGenerator2.htm



Pulse Generating Cells With Relay

luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

48

CI



luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

luxIPlux

LuxI

AHL

48

CI



luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

luxIPlux

LuxI

AHL


48

CI



luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

luxIPlux

LuxI

AHL



48

CI



luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

luxIPlux

LuxI

AHL



Plux({X=cI},…)

48

CI



luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

luxIPlux

LuxI

AHL



Plux({X=cI},…)

…

48

CI



luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

luxIPlux

LuxI

AHL



Plux({X=cI},…)

…

…

48

CI



luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

luxIPlux

LuxI

AHL



Plux({X=cI},…)

…

…

Diff({X=AHL},…)

48

CI



luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPAHL

AHL

luxIPlux

LuxI

AHL



Plux({X=cI},…)

…

…

Diff({X=AHL},…)

Plux({X=luxI},…)

48

CI


Simulation II

Pulse propagation & Rely- simulation II




A Signal Translatorfor Pattern Formation

act1Prep2

act2Prep1

rep1Pact1

rep2Pact2

rep3Prep1

rep4Prep2

I2Prep3

I1Prep4

FP2Pact2

FP1Pact1

50


Uniform Spatial Distribution of Signal Translators for Pattern Formation

51


Pattern Formation in synthetic bacterial colonies

52

Simulation III

http://localhost/~nxk/MOVIES/patternFormation.htm

http://localhost/~nxk/MOVIES/patternFormation.htm


Spatial Distribution of Signal Translators and propagators

luxRPconst

cIPlux

gfpPluxOR1

LuxR GFPSi

Si

luxI

LuxI

Si

Plux

53

CI


Alternating signal pulses in synthetic bacterial colonies

54

Simulation IV

http://localhost/~nxk/MOVIES/alternatingPulses.htm

http://localhost/~nxk/MOVIES/alternatingPulses.htm






•Conclusions

55


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 iterated56


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).

• Jaramillo et al. use Simulated Annealing as the main search strategy for model inference based on (O)DEs

57


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

58

Multi-Objective Optimisation in Morphogenesis


Parameter Optimisation in Metabolic Models

59

A. Drager et al. (2009). Modeling metabolic networks in C. glutamicum: a comparison of rate laws in combination with various parameter optimization strategies. BMC Systems Biol ogy 2009, 3:5


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 evolvability of models/network motifs

60


Nested EA for Model SynthesisF. 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.)

H. Cao, F.J. Romero-Campero, S. Heeb, M. Camara, and N. Krasnogor. Evolving cell models for systems and synthetic biology. Systems and Synthetic Biology, 4(1), 2010.

C. Garcia-Martinez, C. Lima, J. Twycross, M. Lozano, N. Krasnogor. P System Model Optimisation by Means of Evolutionary Based Search Algorithms. Proceedings of the 2010 Genetic and Evolutionary Computation Conference (GECCO-2010). (Nominated for best paper award at the Bioinformatics track.)

61


Fitness Evaluation

62


The Fitness Function• Multiple time-series per target

• Different time series have very different profiles, e.g., maxima occur at different times/places

• Transient states (sometimes) as important as steady states

•RMSE not the only possibility

H. Cao, F.J. Romero-Campero, S. Heeb, M. Camara, and N. Krasnogor. Evolving cell models for systems and synthetic biology. Systems and Synthetic Biology, 4(1), 2010.

63


Some Case Studies

64


Problem Specification

65


Results Study Case 4

66






Target

71


Target

72


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).

Alternative models can be formally analysed and compared based on other than “fitness” criteria:

★Sensitivity analysis / Robustness ★Average properties (model checking)★Parsimony

Comparing results by only using elementary modules or by adding newly found modules to the library shows the obvious advantage of the incremental methodology with modules.

73






•Conclusions

74

BEACON, Michigan State University, July 2010 /8275InfoBiotics Workbench and Dashboard

Spec

ifica

tion

Sim

ulat

ion

Analysis

VerificationManagement

Optim

isation


• P systems are a handy way of specifying discrete and stochastic rule-based compartmental models.

• Modularity in P systems as a design principle for synthetic networks that enables reusability, hierarchical abstraction and standardisation.

• Can be linked to DPD-type simulations. • Model Checking for the formal analysis of our models.

• Automated explorations (evolutionary search) on models’ structure and parameters.

• Computer Aided analysis of modular and alternative designs (e.g. synthetic network functionality).

TAKE HOME MESSAGE

76


An Important Difference with “Normal” Programs: a GP challenge?

•Executable Stochastic P systems are not programs with stochastic behavior

•A cell is a living example of distributed stochastic computing.

•How does this impact GP techniques? can they cope? scale? are they parsimonious?

77

function f1(p1,p2,p3,p4){if (p1<p2) RND print p3 RNDelse RND print p4 RND}

function f1(p1,p2,p3,p4){if (p1<p2) and (rand<0.5) print p3else print p4}


•Three Key Ingredients•Inheritable traits•Variation•Selection Pressure

Crazy Idea: Evolution by Natural Selection as a Search Algorithm

78


•Three Key Ingredients•Inheritable traits•Variation•Selection Pressure

Crazy Idea: Evolution by Natural Selection as a Search Algorithm

78


A More Fundamental Question•Fundamental CS results: Averaged over all problems, all search methods perform equally well (or rather bad).

•So why does Nature use Evolution as its search mechanism? why not other methods?

•In Synt Bio people worry a lot about evolvability (cellularity scale video)

•Can we engineer in vivo new search mechanism? e.g. a biological Variable Neighborhood Search?

Wolpert, D.H., Macready, W.G. (1997), "No Free Lunch Theorems for Optimization," IEEE Transactions on Evolutionary Computation 1, 67

79


Could, e.g., an in vivo VNS “defeat” Evolution? in which environments?

• Perhaps SB rather than trying to stop evolution should try to outsmart/outpace it! 80











Acknowledgements

•Jonathan Blake

•Claudio Lima

•Francisco Romero-Campero

•James Smaldon

•Jamie Twycross

•Karima Righetti

Infobiotics Dashboard, Integrated Environment

Infobiotics Workbench Machine Learning & Optimisation

Infobiotics Modeling & Model Checking

Infobiotics Dissipative Particle Dynamics

Infobiotics Stochastic Simulations

Members of my team working on SB2

EP/E017215/1

EP/D021847/1

BB/F01855X/1

BB/D019613/1

• Prof. M. Camara (CBS, UoN)

• Prof. C. Alexander (Pharmacy , UoN)

• Dr. S. Heeb (CBS, UoN)

• Dr. G. Rampioni (CBS, UoN)

81

Molecular Micro-Biology



Education

Towards a Rapid Model Prototyping Strategy for Systems & Synthetic Biology