aquarium: programmable wetlab - tileliamimeur.com fileaquarium: programmable wetlab tileliamimeur,...

1
AQUARIUM: Programmable Wetlab Tileli Amimeur , Yaoyu Yang, David Younger, Miles Gander, Chris Takahashi, Anthony LaMarca, Jeff Scofield, Dieter Fox, Jeff Bilmes, Eric Klavins University of Washington, Seattle, WA Reproducibility Issues in Wetlabs Aquarium Approach Protocols With ‘Krill’ Workflow Optimization Metacols With ‘Oyster’ Training Current and Future Directions A study in 2012 found that only six of fifty- three selected ‘landmark’ papers in cancer biology were reproducible. Another found that retractions, due to errors or otherwise, are increasingly frequent. Engineers attempting to build on scientific results in the life sciences in order to explore the study of engineering life itself are finding that these results do not hold. Aquarium has two separateservers for protocoldevelopment and in-lab protocol execution. The Rehearsal server is used for testing of protocols/metacols only. The Production server hosts the version of Aquarium thatgets run in the lab by technicians and holds an accurate representation of the lab’s inventory. In order to ensure reproducibility and confidence in Aquarium lab technicians, Aquarium itself provides standardtrainingfor all incoming technicians. A libraryof protocols exists for new tech lab orientation and training and provides lab managers with a way of monitoringand managingtechnicians’ skills. Currently we have a working build pipeline library for E. coli and Yeast , and are in the process of optimizingas manyof these protocols as possible. We are collaborating with the Intel Center for Pervasive Computing to expand Aquarium to add: Tools for protocoldevelopment Object tracking – knowing the exact location of all objects in the lab at all times Automated lab assistance– protocolprogress tracking and error checking, location assistance,physical robotic assistant Future Goals include: Executable code as Materials/Methods: Reproducibilitystandard Networked cloud lab Self-booting lab Lab Technicians Log Database Analysis Interpreter (touch screen) Inventory Manager (for any kind of sample) Squid Manta Lab equipment and devices are managed by the Squid device manager. Manta handles video recording and other sensors. Protocols and Metacols are written in standard languages and stored in version controlled Github repositories. Images are kept in the Angler image server. Researcher Design Build / Test Formal Ruby-based protocol specification Control, branching, etc. Inventory interactions Transfers of liquids between containers Interactions with lab equipment Informal Textual descriptions of steps Photos/cartoons/drawings Krill Protocol Language Lab Technicians Log Database Production Server Rehearsal Server Aquarium is the Klavins lab approachto resolving reproducibility issues in the wetlab. It aims to separate the Design portion of synthetic biology from the Build and Test portions. Aquariumcomes with a language for semi-formalprotocoldescriptions ( Krill ) and a formalprocess controllanguage ( Oyster) for directing the flow of protocols. This was used to build a version controlled Github library of protocols for strain building/testing. The researcher uses Aquarium to select desired protocols/metacols from the Github repo and schedules them to be executed by a trained lab technician. At execution time, the Interpreter displays the protocols as a series of steps on a touch screen for the technician to run through. Data is logged into a database accessible by the researcher. error !error primer_ids plate_id A Metacol specifies process control and is a wiring of protocols and their transitions in the form of Petri Nets Places are jobs associated with a protocol parameters a user group a desired start / stop time Transitions have conditions errors/cancelations inventory status user availability Inputs and outputs can be wired argument gels : sample(“Gel slice”) array fragments : sample(“Fragment” ) array end place purify_gel protocol : ”purify_gel.pl" argument gel : gel end group : aquarium_user marked : true start : now() window : minutes(30 ) end wire (purify_gel,”fragment_stock_ids" ) => (gibson_prepare, "fragments" ) # forward transitions transition [purify_gel] => [dpnI_digest ] when !error(0) && completed (0) end transition [dpnI_digest ] => [pcr_purification ] when !error(0) && completed (0) end transition [pcr_purification] => [gibson_prepare] when !error(0) && completed (0) end transition [gibson_prepare ] => [stock_fragment] when !error(0) && completed (0) end transition [stock_fragment ] => [] when !error(0) && completed (0) end Oyster Metacol Language PCR/Gels Purify Fragments Gibson Transform Miniprep Sequence Pipeline for Building a New Strain Transform Plate Incubate Image Electrocompetent cells Liquid Media Solid Media Plate Inventory Samples Inventory Objects Associated Watchers In order to keep the pipeline flowing, all necessary input parameters need to be available(dependencies need to be resolved). We have what is called a watcher that monitors key inventory samples and objects and fires restockingprotocols when an item runs low. Quality Checks Optimization Buildinga new logic function into a cell involves stepping through a complex set of highly variablewetlab cloning protocols. A main portion of the build pipeline is shown above. One place for high variabilityis the Transform step. Transformation involves electroshockingthe cells enough to temporarily weaken their cell walls to allow acceptance of the new plasmid, which has the desired logic function encoded onto it. The set of protocols necessary for executing and testingthe transformationare shown here on the right. In order to minimize variability in the build process, we introduce quality checks to test crucial points in the pipeline. There are numerous places in the build protocols where parameters can be optimized. For example, the temperature at which certain chemicals are added, the length of time at which they’re mixed, the concentration of plasmid mixture to grow up. We can tease out the ideal values for these parameters by runningvariations of protocols and extractingthe values that give us the best results. In addition, with Aquarium’s datacollection methods, we can conceivably reveal experimentalrelationships between parameters that we didn’t know existed before. Wait for low comp cell stock Make comp cells QC: Test comp cells Discard batch Electrocompetent cells Pass Fail In addition to tech training, we’ve used Aquarium to teach synthetic biology to graduatestudents from all backgrounds.In five weeks students learned basic lab skills, molecular cloning, Gibson assembly, and gene knockouts with very little external assistanceoutsideof Aquarium. Protocol/Object Tracking Robotic Assistance Even when a result is valid, the methods are poorly explained, purposefully not explained, or simply buried in some researcher’s notes. Science suffers from a growing need to be able to specify, as precisely as possible, the means for obtaining a result. A standard language for reproducible methods.

Upload: others

Post on 31-Aug-2019

6 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: AQUARIUM: Programmable Wetlab - tileliamimeur.com fileAQUARIUM: Programmable Wetlab TileliAmimeur, Yaoyu Yang, David Younger, Miles Gander, Chris Takahashi, Anthony LaMarca, Jeff Scofield,

AQUARIUM:ProgrammableWetlabTileli Amimeur,Yaoyu Yang,DavidYounger,MilesGander,ChrisTakahashi,AnthonyLaMarca,

JeffScofield,DieterFox,JeffBilmes,EricKlavinsUniversityofWashington,Seattle,WA

ReproducibilityIssuesinWetlabs

AquariumApproach

ProtocolsWith‘Krill’ WorkflowOptimization

Metacols With‘Oyster’

Training

CurrentandFutureDirections

Astudyin2012foundthatonlysixoffifty-threeselected‘landmark’papersincancerbiologywerereproducible.

Anotherfoundthatretractions,duetoerrorsorotherwise,areincreasinglyfrequent.Engineersattemptingtobuildonscientificresultsinthelifesciencesinordertoexplorethestudyofengineeringlifeitselfarefindingthattheseresultsdonothold.

Aquariumhastwoseparateserversforprotocoldevelopmentandin-labprotocolexecution.TheRehearsalserverisusedfortestingofprotocols/metacols only.TheProductionserverhoststheversionofAquariumthatgetsruninthelabbytechniciansandholdsanaccuraterepresentationofthelab’sinventory.

InordertoensurereproducibilityandconfidenceinAquariumlabtechnicians,Aquariumitselfprovidesstandardtrainingforallincomingtechnicians.Alibraryofprotocolsexistsfornewtechlaborientationandtrainingandprovideslabmanagerswithawayofmonitoringandmanagingtechnicians’skills.

CurrentlywehaveaworkingbuildpipelinelibraryforE.coli andYeast,andareintheprocessofoptimizingasmanyoftheseprotocolsaspossible.

WearecollaboratingwiththeIntelCenterforPervasiveComputingtoexpandAquariumtoadd:

•Toolsforprotocoldevelopment•Objecttracking– knowingtheexactlocationofallobjectsinthelabatalltimes•Automatedlabassistance– protocolprogresstrackinganderrorchecking,locationassistance,physicalroboticassistant

FutureGoals include:

•ExecutablecodeasMaterials/Methods:Reproducibilitystandard•Networkedcloudlab•Self-bootinglab

Researcher

LabTechnicians LogDatabase Analysis

Interpreter(touchscreen)

InventoryManager

(foranykindofsample)

Squid

Manta

LabequipmentanddevicesaremanagedbytheSquiddevicemanager.

Mantahandlesvideorecordingandother

sensors.

ProtocolsandMetacols arewritteninstandard languagesandstoredin

versioncontrolledGithubrepositories. Imagesarekept inthe

Anglerimageserver.

Researcher Design

Build / Test

Formal• Ruby-based protocol specification• Control, branching, etc.• Inventory interactions• Transfers of liquids between containers• Interactions with lab equipment

Informal• Textual descriptions of steps• Photos/cartoons/drawings

KrillProtocolLanguage

LabTechnicians

LogDatabase

ProductionServer

RehearsalServer

Aquarium istheKlavinslabapproachtoresolvingreproducibilityissuesinthewetlab.ItaimstoseparatetheDesignportionofsyntheticbiologyfromtheBuildandTestportions.Aquariumcomeswithalanguageforsemi-formalprotocoldescriptions(Krill)andaformalprocesscontrollanguage(Oyster)fordirectingtheflowofprotocols.ThiswasusedtobuildaversioncontrolledGithub libraryofprotocolsforstrainbuilding/testing.

TheresearcherusesAquariumtoselectdesiredprotocols/metacols fromtheGithub repoandschedulesthemtobeexecutedbyatrainedlabtechnician.Atexecutiontime,theInterpreterdisplaystheprotocolsasaseriesofstepsonatouchscreenforthetechniciantorunthrough.Dataisloggedintoadatabaseaccessiblebytheresearcher.

error

!error

primer_ids

plate_id

• AMetacol specifiesprocesscontrolandisawiringofprotocolsandtheirtransitionsintheformofPetriNets

• Places arejobsassociatedwith• aprotocol• parameters• ausergroup• adesiredstart/stoptime

• Transitions haveconditions• errors/cancelations• inventorystatus• useravailability

• Inputs andoutputs canbewired

argumentgels : sample(“Gel slice”) arrayfragments : sample(“Fragment”) array

end

place purify_gelprotocol : ”purify_gel.pl"argument

gel : gelendgroup : aquarium_usermarked : truestart : now()window : minutes(30)

end

wire (purify_gel,”fragment_stock_ids") => (gibson_prepare,"fragments")

# forward transitionstransition [purify_gel] => [dpnI_digest] when !error(0) && completed(0) endtransition [dpnI_digest] => [pcr_purification] when !error(0) && completed(0) endtransition [pcr_purification] => [gibson_prepare] when !error(0) && completed(0) endtransition [gibson_prepare] => [stock_fragment] when !error(0) && completed(0) endtransition [stock_fragment] => [] when !error(0) && completed(0) end

OysterMetacol Language

PCR/Gels PurifyFragments Gibson Transform Miniprep Sequence

PipelineforBuildingaNewStrain

Transform

Plate

Incubate

Image

ElectrocompetentcellsLiquidMedia

SolidMediaPlate

InventorySamples

InventoryObjects

AssociatedWatchersInordertokeepthepipelineflowing,allnecessaryinputparametersneedtobeavailable(dependenciesneedtoberesolved).Wehavewhatiscalledawatcherthatmonitorskeyinventorysamplesandobjects andfiresrestockingprotocolswhenanitemrunslow.

QualityChecks

Optimization

Buildinganewlogicfunctionintoacellinvolvessteppingthroughacomplexsetofhighlyvariablewetlab cloningprotocols.Amainportionofthebuildpipelineisshownabove.OneplaceforhighvariabilityistheTransform step.Transformationinvolveselectroshockingthecellsenoughtotemporarilyweakentheircellwallstoallowacceptanceofthenewplasmid,whichhasthedesiredlogicfunctionencodedontoit.Thesetofprotocolsnecessaryforexecutingandtestingthetransformationareshownhereontheright.

Inordertominimizevariabilityinthebuildprocess,weintroducequalitycheckstotestcrucialpointsinthepipeline.

Therearenumerousplacesinthebuildprotocolswhereparameterscanbeoptimized.Forexample,thetemperatureatwhichcertainchemicalsareadded,thelengthoftimeatwhichthey’remixed,theconcentrationofplasmidmixturetogrowup.Wecanteaseouttheidealvaluesfortheseparametersbyrunningvariationsofprotocolsandextractingthevaluesthatgiveusthebestresults.Inaddition,withAquarium’sdatacollectionmethods,wecanconceivablyrevealexperimentalrelationshipsbetweenparametersthatwedidn’tknowexistedbefore.

Waitforlowcompcellstock

Makecompcells

QC:Testcompcells

Discardbatch

Electrocompetentcells

PassFail

Inadditiontotechtraining,we’veusedAquariumtoteachsyntheticbiologytograduatestudentsfromallbackgrounds.Infiveweeksstudentslearnedbasiclabskills,molecularcloning,Gibsonassembly,andgeneknockoutswithverylittleexternalassistanceoutsideofAquarium.

Protocol/ObjectTracking

RoboticAssistance

Evenwhenaresultisvalid,themethodsarepoorlyexplained,purposefullynotexplained,orsimplyburiedinsomeresearcher’snotes.Sciencesuffersfromagrowingneedtobeabletospecify,aspreciselyaspossible,themeansforobtainingaresult.Astandardlanguageforreproduciblemethods.