Rule based modelling symmetries refinements

Vincent Danos1

Description

Niveau: Supérieur, Doctorat, Bac+8
Rule-based modelling, symmetries, refinements Vincent Danos1,3, Jerome Feret2, Walter Fontana2, Russell Harmer3, and Jean Krivine2 1 University of Edinburgh 2 Harvard Medical School 3 CNRS, Universite Paris Diderot Abstract. Rule-based modelling is particularly effective for handling the highly combinatorial aspects of cellular signalling. The dynamics is de- scribed in terms of interactions between partial complexes, and the ability to write rules with such partial complexes -i.e., not to have to specify all the traits of the entitities partaking in a reaction but just those that matter- is the key to obtaining compact descriptions of what otherwise could be nearly infinite dimensional dynamical systems. This also makes these descriptions easier to read, write and modify. In the course of modelling a particular signalling system it will often happen that more traits matter in a given interaction than previously thought, and one will need to strengthen the conditions under which that interaction may happen. This is a process that we call rule refinement and which we set out in this paper to study. Specifically we present a method to refine rule sets in a way that preserves the implied stochastic semantics. This stochastic semantics is dictated by the number of different ways in which a given rule can be applied to a system (obeying the mass ac- tion principle). The refinement formula we obtain explains how to refine rules and which choice of refined rates will lead to a neutral refinement, i.

• rule-based modelling


• refinement


• interaction


• micro- causal constraints


• rewriting framework


• refinement neutral


• single site


• partial complexes


• interactions than

Sujets

• Refinement
• Interaction

Informations

• Publié par: profil-zyak-2012
• Nombre de lectures: 21
• Langue: English

Extrait

Rule-basedmodelling,symmetries,refinements VincentDanos 1,3 ,Je´roˆmeFeret 2 ,WalterFontana 2 ,RussellHarmer 3 ,andJean Krivine 2 1 UniversityofEdinburgh 2 HarvardMedicalSchool 3 CNRS,Universite´ParisDiderot Abstract.Rule-basedmodellingisparticularlyeffectiveforhandlingthe highlycombinatorialaspectsofcellularsignalling.Thedynamicsisde- scribedintermsofinteractionsbetweenpartialcomplexes,andtheability towriteruleswithsuchpartialcomplexes-i.e.,nottohavetospecify allthetraitsoftheentititiespartakinginareactionbutjustthosethat matter-isthekeytoobtainingcompactdescriptionsofwhatotherwise couldbenearlyinfinitedimensionaldynamicalsystems.Thisalsomakes thesedescriptionseasiertoread,writeandmodify. Inthecourseofmodellingaparticularsignallingsystemitwilloften happenthatmoretraitsmatterinagiveninteractionthanpreviously thought,andonewillneedtostrengthentheconditionsunderwhichthat interactionmayhappen.Thisisaprocessthatwecallrulerefinement andwhichwesetoutinthispapertostudy.Specificallywepresenta methodtorefinerulesetsinawaythatpreservestheimpliedstochastic semantics. Thisstochasticsemanticsisdictatedbythenumberofdifferentways inwhichagivenrulecanbeappliedtoasystem(obeyingthemassac- tionprinciple).Therefinementformulaweobtainexplainshowtorefine rulesandwhichchoiceofrefinedrateswillleadtoaneutralrefinement, i.e.,onethathasthesameglobalactivityastheoriginalrulehad(and thereforeleavesthedynamicsunchanged).Ithasapleasingmathematical simplicity,andisreusablewithlittlemodificationacrossmanyvariantsof stochasticgraphrewriting.Aparticularcaseoftheaboveisthederivation ofamaximalrefinementwhichisequivalenttoa(possiblyinfinite)Petri netandcanbeusefultogetaquickapproximationofthedynamicsand tocalibratemodels.Asweshowwithexamples,refinementisalsouseful tounderstandhowdifferentsubpopulationscontributetotheactivityof arule,andtomodulatedifferentiallytheirimpactonthatactivity. 1Semi-liquidcomputing Totheeyeofthecomputationalscientist,cellularsignallinglookslikeanintrigu- ingcomputationalmedium.Varioustypesofagents(proteins)oflimitedmeans interactinwhat,atfirstsight,mayseemtobealiquiduniverseofchanceencoun- terswherethereislittlecausality.Butinfactarichdecentralizedchoreography ofbindings(complexformation)andmutualmodifications(post-translational

modifications)canbeobserved.Transientdevices(complexes)arebuiltbyagents tointegrate,convey,andamplifysignalsandchannelthemtotheappropriate outputs(transcriptionalregulation).Theintricatepathwaysoftheresponseto theepidermalgrowthfactor(EGF)sketchedinFig.1areawell-studiedandwell- modelledexample[1].Thisuniverseofsemi-liquidcomputingisbroughtabout byasurprisinglysmallnumberofelementaryinteractions.Itsitssomewherein betweentheworldsoftherandomgraphsofstatisticalphysics[2]whichperhaps lackexpressivity,andthesolidcollidingspheremodelsofchemicalkinetics[3] whichperhapslackprogrammability. Thegenerativityofthosesystems,thatistosaythenumberofdifferent non-isomorphiccombinations(akacomplexesorspecies)thatmaycometoexist alongdifferentrealizationsoftheimpliedstochasticprocess,maywellbeenor- mous,butthisdoesnotsayhowcomplexthosesystemsreallyare.Alotfewer rulesthantherearereactions(interactionsbetweencompletecomplexes)may begoodenoughtodescribesomeofthem.ForinstancethesketchofFig.1once properlyformalizedusesabout300ruleswhereasitproducesabout10 40 unique combinations.Oneseesthatthenumberofrulesisamoremeaningfulestimate ofitsinherentcomplexity. Rule-basedlanguages[4–11],andmoregenerallyprocessalgebraicapproaches tomodelling[12–19],becausetheycanexpresssuchgenericinteractions,can workaroundthisapparentdescriptivecomplexityandachievecompactdescrip- tions.Letusalsomention,althoughwewillnottreatthisaspectofthequestion here,thatanotherbenefitofrule-basedmodellingisthatonecantracetheevo- lutionofasystematthelevelofagents(orindividuals)andexplorethecausal relationshipsbetweeneventsoccurringinasystem[6]. Thedifferencebetweenanassemblyofagentswithrandomuncorrelateden- countersandasignallingsystemisthatthereisacausalstructurechannellingthe interactionstowardsaparticularresponse.Typicallyabindingwillnothappen beforeoneorbothofthebindeeshasbeenmodified.Combiningthosemicro- causalconstraintsintoacoherentpathwayisaprogrammingartthatwedon’t masterorevenunderstandyet,butonethatsignallingsystemshavebeenhoning foraconsiderabletime.Rule-basedmodellingincorporatessuchcausalitycon- straintsintherulesthemselvesbyusingpartialcomplexes:noteverythingneeds tobedescribedinarule,onlytheaspectsofthestateofacomplexwhichmatter foraneventtohappenneedtobespecified.Thatisthedifferencebetweena reactionbetweencompleteentities,andarulebetweenpartialones.Assaid,this relianceonpartialcomplexesallowstocapturecompactdescriptionsandwork aroundthehugenumbersofcombinationsonewouldhavetocontemplate(or neglect)otherwise.Themoredetailedthepartialcomplex,thatistosaytheless partial,themoreconditionsmustbemetforaparticulareventtohappen. Thepurposeofthepresentpaperistounderstandbetterthemechanicsof refinement,thatistosaytheprocessbywhichonecanmakeacomplexless partial,orequivalentlyarulemoredemanding.Wespecificallyconsiderthe problemofreplacingarulewithafamilyofrefinedruleswhichwillexhibit thesamecollectiveactivity,andwillthereforegenerateanidenticalstochastic 2