RESEARCH ARTICLE SUMMARY◥

CHEMISTRY

Organic synthesis in a modularrobotic system driven by a chemicalprogramming languageSebastian Steiner, Jakob Wolf, Stefan Glatzel, Anna Andreou, Jarosław M. Granda,Graham Keenan, Trevor Hinkley, Gerardo Aragon-Camarasa, Philip J. Kitson,Davide Angelone, Leroy Cronin*

INTRODUCTION:Outside of a few well-definedareas such as polypeptide and oligonucleotidechemistry, the automation of chemical syn-thesis has been limited to large-scale bespokeindustrial processes, with laboratory-scale anddiscovery-scale synthesis remaining predom-inantly a manual process. These areas aregenerally defined by the ability to synthesizecomplex molecules by the successive iterationof similar sets of reactions, allowing the syn-thesis of products by the automation of a rela-tively small palette of standardized reactions.Recent advances in areas such as flow chem-istry, oligosaccharide synthesis, and iterativecross-coupling are expanding the number ofcompounds synthesized by automated meth-ods. However, there is no universal and inter-operable standard that allows the automationof chemical synthesis more generally.

RATIONALE:We developed a standard ap-proach that mirrors how the bench chemistworks and how the bulk of the open literature

is reported, with the round-bottomed flask asthe primary reactor. We assembled a relativelysmall array of equipment to accomplish a widevariety of different syntheses, and our abstrac-tion of chemical synthesis encompasses thefour key stages of synthetic protocols: reaction,workup, isolation, and purification. Further,taking note of the incomplete way chemicalprocedures are reported, we hypothesized thata standardized format for reporting a chemicalsynthesis procedure, coupled with an abstrac-tion and formalism linking the synthesis tophysical operations of an automated roboticplatform, would yield a universal approach to achemical programming language. We call thisarchitecture and abstraction the Chemputer.

RESULTS:For the Chemputer system to accom-plish the automated synthesis of target mole-cules, we developed a program, the Chempiler,to produce specific, low-level instructions formodular hardware of our laboratory-scale syn-thesis robot. The Chempiler takes information

about the physical connectivity and compo-sition of the automated platform, in the formof a graph using the open-source GraphMLformat, and combines it with a hardware-independent scripting language [chemicalassembly (ChASM) language], which providesinstructions for the machine operations of theautomated platform. The Chempiler softwareallows the ChASM code for a protocol to berun without editing on any unique hardware

platform that has the cor-rect modules for the syn-thesis. Formalization of awritten synthetic schemeby using a chemical de-scriptive language (XDL)eliminates the ambiguous

interpretation of the synthesis procedures.This XDL scheme is then translated into theChASM file for a particular protocol. An auto-mated robotic platform was developed, con-sisting of a fluidic backbone connected to aseries of modules capable of performing theoperations necessary to complete a syntheticsequence. The backbone allows the efficienttransfer of the required chemicals into andout of any module of the platform, as well asthe flushing and washing of the entire systemduring multistep procedures in which the mod-ules are reused multiple times. The modulesdeveloped for the system consist of a reac-tion flask, a jacketed filtration setup capa-ble of being heated or cooled, an automatedliquid-liquid separation module, and a sol-vent evaporation module. With these fourmodules, it was possible to automate the syn-thesis of the pharmaceutical compounds di-phenhydramine hydrochloride, rufinamide,and sildenafil without human interaction, inyields comparable to those achieved in tradi-tional manual syntheses.

CONCLUSION: The Chemputer allows for anabstraction of chemical synthesis,when coupledwith a high-level chemical programming lan-guage, to be compiled by our Chempiler intoa low-level code that can run on a modularstandard robotic platform for organic syn-thesis. The software and modular hardwarestandards permit synthetic protocols to becaptured as digital code. This code can bepublished, versioned, and transferred flex-ibly between physical platformswith nomodi-fication. We validated this concept by theautomated synthesis of three pharmaceuticalcompounds. This represents a step towardthe automation of bench-scale chemistrymore generally and establishes a standardaiming at increasing reproducibility, safety,and collaboration.▪

RESEARCH

Steiner et al., Science 363, 144 (2019) 11 January 2019 1 of 1

The list of author affiliations is available in the full article online.*Corresponding author. Email: lee.cronin@glasgow.ac.ukCite this article as S. Steiner et al., Science 363, eaav2211(2019). DOI: 10.1126/science.aav2211

Abstraction oforganic synthesis

Chemical programminglanguage

Modular robotic platform

Digital synthesis

O 2S N

NMe

N

HNEtO

O

N

MeN

nPr

F

F

NN

N

ONH 2

ON

·HCl

Chemputer machine code0: [valve_pos, valve_solvents, flask_product1]

[pump, pump_solvents, 100]1: [valve_pos, valve_solvents, valve_extra] [valve_pos, valve_extra, valve_solvents] [valve_pos, valve_link, valve_extra]

Reaction

Work-up

Isolation

Purification

Filter

Separator

Reactor

Evaporator

Sildenafil

Rufinamide

NytolThe Chemputer

GraphML codeChASM code

Abstraction of organic synthesis to the Chemputer software and hardware standards.The Chemputer system can perform multistep organic synthesis protocols on the bench byusing a modular system with hardware and software standards.

ON OUR WEBSITE◥

Read the full articleat http://dx.doi.org/10.1126/science.aav2211..................................................

on February 16, 2020

http://science.sciencem

ag.org/D

ownloaded from

RESEARCH ARTICLE◥

CHEMISTRY

Organic synthesis in a modularrobotic system driven by a chemicalprogramming languageSebastian Steiner1*, Jakob Wolf1, Stefan Glatzel1, Anna Andreou1,Jarosław M. Granda1, Graham Keenan1, Trevor Hinkley1, Gerardo Aragon-Camarasa1,2,Philip J. Kitson1, Davide Angelone1, Leroy Cronin1†

The synthesis of complex organic compounds is largely a manual process that is oftenincompletely documented. To address these shortcomings, we developed an abstractionthat maps commonly reported methodological instructions into discrete steps amenableto automation. These unit operations were implemented in a modular robotic platformby using a chemical programming language that formalizes and controls the assemblyof the molecules. We validated the concept by directing the automated system tosynthesize three pharmaceutical compounds, diphenhydramine hydrochloride, rufinamide,and sildenafil, without any human intervention. Yields and purities of products andintermediates were comparable to or better than those achieved manually. The synthesesare captured as digital code that can be published, versioned, and transferred flexiblybetween platforms with no modification, thereby greatly enhancing reproducibility andreliable access to complex molecules.

The automation of chemical synthesis is cur-rently expanding, and this is driven by theavailability of digital labware. The field cur-rently encompasses areas as diverse as thedesign of new reactions (1), chemistry in

reactionware (2), reaction monitoring and opti-mization (3, 4), flow chemistry (5) for reactionoptimization and scale up, and even full automa-tion of the synthesis laboratory (6). Establishedtechnologies such as the automated synthesis ofpeptides (7) and oligonucleotides (8), flow chem-istry (9), and high-throughput experimentation(6) are mainstays of modern chemistry, whereasemerging technologies such as automatedoligosac-charide synthesis (10) and iterative cross-coupling(1) have the potential to further transform thechemical sciences. Each of these examples, how-ever, relies on a distinct protocol, as no currentdigital automation standard exists for computercontrol of chemical reactions (11). Hence, no gen-eral programming language is available for chem-ical operations that can direct the synthesis oforganic compounds on an affordable, flexible,modular platform accessible to synthetic chem-ists and that could, in principle, encompass allsynthetic organic chemistry. This situation is com-parable to the era before digital programmablecomputers,when existing computational deviceswere fixed to a dedicated problem.

A generalized approach to automating chem-ical synthesiswould be beneficial becausemakingcompounds is one of the most labor-intensivebranches of chemistry, requiringmanual execu-tion of a range of unit operations such as reagentmixing, liquid-liquid extractions, or filtrations.Despite the modular nature of the operations,standardization and automation are still severe-ly limited. Furthermore, the ambiguous way inwhich synthetic protocols are communicated hascontributed to a mounting reproducibility crisis(12). Syntheses are reported in prose, omittingmany details explaining exactly how operationswere carried out and making many assumptionsabout the skill level of the researcher repeatingthe process. We hypothesized that a more stan-dardized format for reporting a chemical syn-thesis procedure, coupled with an abstractionand formalism linking the synthesis to thephysical operations, could yield a universal ap-proach to a chemical programming language;we call this architecture and abstraction theChemputer.In developing the Chemputer platform, we

wanted to build on the 200 or more years ofchemical literature and the experience of themany thousands of bench chemists active in theworld today in a way that would naturally leadto a standardization embodied in a codified stan-dard recipe, or chemical code, for molecular syn-thesis. For this to be possible, it was essential thatthe approach mirror the way the bench chemistworks. As our key reaction module for imple-mentation of the Chemputer, we selected theround-bottomed flask for batch synthesis, with

well-defined inputs and outputs, because mostprotocols already published for complexmoleculesynthesis rely on this type of apparatus. Next, weidentified the four key stages of a synthetic pro-tocol from our abstraction: reaction, workup,isolation, and purification. These stages can besubdivided into several unit operations, which inturn are implemented in a specific automatedplatform.

Developing code for chemistry

Bydeveloping control software aswell as hardwaremodules for laboratory-scale synthesis that canbe automatically cleaned and reused in subse-quent reaction steps, we were able to define aprocess for combining individual unit opera-tions into full, multistep syntheses to producedesired products autonomously (Fig. 1A). For theChemputer to operate as shown in Fig. 1B, thestates of the inputs, reactor, and outputs mustbe defined and controlled programmatically.Wetherefore created a Chempiler, which is a pro-gram to produce specific low-level instructionsfor the modules that constitute the Chemputerarchitecture. The Chempiler can run commandsused to control the modular platform from ourabstraction layer, so that a typical written schemecan be turned into a specific code to run themod-ules with ease. Every module was then designedwith drivers for the device or equipment and astandardized application programming inter-face (API) exposing the instruction set to theChempiler (Fig. 2). The use of a program coupledto the Chemputer architecture allows users todirectly run published syntheses without recon-figuration, provided the necessary modules anddrivers are present within their system. Thus,the practical implementation of the Chemputerarchitecture converts digital code to chemicalsynthesis operations in accordance with thestandard protocol of a chemical reaction incor-porating the four processes listed in Fig. 1C. Toprevent the need for manual reconfiguration,the physical modules and their connections andrepresentation are stored inmemory as a directedgraph, which allows knowledge and control oftheir states in real time.The physical routing that links the connected

modules is described in a graph by using anopen-source format, GraphML, which allows theChempiler to find paths between a source flaskand a target flask, as well as address devices suchas hot plate stirrers on the basis of the vessel theyare connected to. GraphML is an open-standard,extensible markup language (XML)–based ex-change format for graphs (13) [example graphsused in the syntheses described herein can befound in data S1 (GraphML files) in the supple-mentary materials (SM)]. Synthetic proceduresare codified by using a scripting language calledChemical Assembly (ChASM), which providesinstructions for all currently implemented ma-chine operations and supports the definition offunctions and variables. To develop the ChASMcode,we built a standard procedure, startingwitha written synthetic scheme that is formalizedby using a chemical descriptive language (ΧDL).

RESEARCH

Steiner et al., Science 363, eaav2211 (2019) 11 January 2019 1 of 8

1School of Chemistry, The University of Glasgow, GlasgowG12 8QQ, UK. 2School of Computer Science, The Universityof Glasgow, Glasgow G12 8QQ, UK.*Present address: Department of Chemistry, University of BritishColumbia, Vancouver, British Columbia V6T 1Z1, Canada.†Corresponding author. Email: lee.cronin@glasgow.ac.uk

on February 16, 2020

http://science.sciencem

ag.org/D

ownloaded from

The XDL has the advantage of eliminatingambiguous interpretation of the synthesis pro-cedures by explicitly and systematically listingall required information without making anyassumptions or inferences (example ChASMfiles used in the syntheses described hereincan be found in data S2).To control the chemistry, the Chempiler was

designed to accept ChASM commands such as“start stirring the reactor,” find the module inquestion in the GraphML definition, and sched-ule the execution by using the appropriate low-level instructions. The modular Chemputercomponents thereby constitute a versatile andinteroperable architecture for chemical synthesis.Also, a given ChASM file could be run on manydifferent platform instances with hardware ofdifferent makes and models connected in dif-ferent ways. If the required unit operationmod-ules are present and all required reagents andsolvents are provided, the hardware-agnosticChASM code can be freely paired with a system-specific GraphML file to synthesize the same

product on another, different platform withoutreoptimization.During the development of the process and

programming language, we found it helpful tovisualize the workflow one would follow whenmanually reproducing the procedure. Writingthe XDL and converting it to ChASM then be-come intuitive even for users with no program-ming experience. Many operations are repetitiveenough to be defined as functions that can be re-used many times. For instance, we have definedseveral functions such as “evaporate to dryness”or “add reagents and heat up to x°C,” and weexpect that this library can be greatly expandedwith future use. Once the ChASM file is com-pleted, a graph of the platform is drafted, follow-ing a set of rules detailed in the SM. To validatethe ChASM and the graph, we implemented twosimulation modes in the Chempiler. First, alloperations are performed as they would be onthe real system, but instead of commands beingissued to physical devices, the commands arelogged to a text file. Next, the system simulates

the process without reagents as a dry run. Thesimulations flag any potential issues, such assyntax errors in the XDL representation and theChASM, inconsistencies in the graph represen-tation, impossible operations, or overfilling ofvessels. Once the simulations run without errors,the user can load the reagents and solvents inaccordance with the graph and start the syn-thesis. The Chempiler also has a break-pointcommand option for additional safety and com-patibility checks as needed.

Modular synthesis platform

To produce a physical platform that could imple-ment our architecture and achieve the synthesesoutlined below, we had to step away from think-ing about reactions and rather adopt a process-centered way of thinking. Although the 20 mostcommonly used reaction classes in drug discov-ery (14) span a wide range of chemistries, fromamide bond formation to Buchwald-Hartwigcouplings, experimentally most of them simplyinvolve the mixing of several reagents in the

Steiner et al., Science 363, eaav2211 (2019) 11 January 2019 2 of 8

Abstraction, Processes, and Programming of Chemical Synthesis

Pump OperationsMOVE ({src},{dest}…)PRIME ({speed}…)SEPARATE ({phase,…)

Stirring OperationsSTART_STIR ({name})STOP_STIR ({name})SET_STIR_RPM ({rpm})

Heating OperationsSTART_HEAT ({name})STOP_HEAT ({name})SET_TEMP ({temp},…)

Vacuum OperationsSTART_VAC ({name})SET_VAC_SP ({name})STOP_VAC ({name},…)

Cooling OperationsSTART_CHILLER ({name})STOP_CHILLER ({name})SET_TEMP ({temp},…)

ReactionAbstraction Chemical Processes

Add Reagents

Heating / Cooling

Quench Reaction

Filtration

Biphasic Extraction

Crystallization

Evaporation

Mixing

Clean and Prime System

Transfer Between Modules

Chemputer Architecture

System Bus

I/O ModuleASM, Assembler,

Buffer

ChemPUALU, Control Units,

Chemical Logic Unit,Registers

MemoryInstructions, Data

GraphML

DigitalMemory

Pumps, Valves,Chemicals

PhysicalPhysical Bus

A B NC

A product B

N

product CB

Chemical Processing Unit

Physical Memory:

Input:Starting

Materials

Reaction

Work-up

Isolation

Purification

Machine Operations

N-Step Chemical synthesis

Output:Final

Product

A

Hardware Operations

Pumps, valves,Chemical Shift Registers

Physical I/O Module

F

F

H

OBr

A B

C

Fig. 1. Operating principles of the Chemputer. (A) Schematic representation of a stepwise chemical synthesis, formalizing the reagents as inputsand products as output. (B) Diagram outlining the computing architecture of the Chemputer. ALU, arithmetic logic unit; I/O, input/output; ASM,assembly language. (C) The abstraction of chemical synthesis allowing the development of an ontology that can be universally programmed by usinga machine. Similar to digital computers, the Chemputer has a memory and a bus system, but these are both digital and physical. By consideringchemical reagents and products in a memory bus, it is possible to break the process of complex molecule synthesis into steps or cycles that can berun using the physical hardware.

RESEARCH | RESEARCH ARTICLEon F

ebruary 16, 2020

http://science.sciencemag.org/

Dow

nloaded from

correct order, often under heating or cooling.These processes are typically followed by aworkup and a purification technique such asdistillation, recrystallization, or chromatogra-phy. We therefore concluded that a synthesisplatform capable of performing the unit oper-ations of mixing under heating or cooling,liquid-liquid separation, filtration, and solventevaporation could in principle perform a largefraction of all organic syntheses and embodyour abstraction of chemical synthesis (Fig. 1C).As those unit operations do not always occurin the same order, especially where multistepsyntheses are concerned, a flexible means ofmoving material between the modules was re-quired. To that end, we built the physical archi-tecture around a fluidic backbone consisting ofa series of syringe pumps and six-way selection

valves (Fig. 3). The appeal of this design is thatit is expandable: The user can always add morepump-valve units (backbone units) to the endsof the backbone. Material can bemoved betweenmodules in an arbitrary order, includingmultipleuses of the same module at different points inthe sequence. The process to transfer liquid froma port on the backbone to another backbone isdescribed in the SM (fig. S8). The pump on thesource unit aspirates the appropriate amount,then the valve on the source unit and the adja-cent unit switch to the bridge, and the sourcepump and the adjacent pump move simulta-neously to transfer the liquid contents from thesource syringe to the next syringe. This processis repeated, and the liquid is moved along thebackbone until it reaches the destination unit,which in turn dispenses it to the destination port.

Surplus ports on the six-way valves accommodatesolvents and reagents, and additional backboneunits may be dedicated entirely to supplyingchemicals. This is in contrastwith even themostadvanced flow chemical setups to date (15), whichhave to be physically reconfigured for every newsynthesis. In flow, the number and sequence ofunit operation modules must match the num-ber and sequence of required unit operations,whereas in our system the ability to addressmod-ules independently and reuse them as requiredremoves the need for physical reconfiguration.This is achieved by using GraphML, which givesa complete description of the connectivity, al-lowing the pumps and valves to be dynamicallyreconfigurable resources, and facilitates themovement of solvents and solutions from re-agent flasks to the various components requiredfor a given synthesis step. To ensuremodularity,we designed our own pumps and valves, con-trolled and powered by a single Ethernet cableplugged into each item, powered from a networkswitch placed next to the fume hood (see fig. S11).In this work, we developed modules for the

unit operations ofmixing, filtration, liquid-liquidseparation, evaporation, and chromatographicseparation, which are all key to the practicalimplementation of our abstraction of a chemicalreaction. Detailed descriptions of the individualmodules can be found in figs. S34 to S44 andaccompanying supplementary text. However,both the physical architecture and the architec-ture of the Chempiler (vide supra) were specif-ically designed to allow for future addition ofother modules, enabling automation of evenmore reactions.The reactor (fig. S34) consisted of a commer-

cially available, two-necked, pear-shaped flaskequipped with an air condenser. We decided touse common laboratory glassware rather thana jacketed reactor vessel both for simplicity andto lower the barrier for reproducing the setup.A pear-shaped flask was chosen over the morecommon round-bottomed flask to accommodatea wider range of reaction volumes. Heating andstirringwere accomplished by using a computer-controllable stirrer hot plate and a custom-manufactured aluminum heating block for thepear-shaped flask. A 0.32-cm-outer-diameterpolytetrafluoroethylene (PTFE) tube was heldin place by a ground glass joint–to–GL18 threadadapter with a GL18 screw cap and insert so thatits end reached the bottom of the flask. A slightargon overpressure was maintained by the inertgas system (see SM). For the sildenafil synthesis,cooling of the reactor was required, andwe useda recirculation chiller with a temperature rangeof −30 to 160°C, giving precise temperature con-trol of the reactor.The liquid-liquid separation, one of the most

common isolation techniques, was also themostchallenging task to automate in a robust fashion.Whereas in continuous flow there are solutionsusing membrane technology (16), we found thatcommercially available hydrophobic frits areusually designed to be single use and thereforelack long-term reliability. We attempted to use

Steiner et al., Science 363, eaav2211 (2019) 11 January 2019 3 of 8

Fig. 2. Diagram showing the flow of information in the synthesis platform. To translateexperimental steps into a set of pump movements or hot plate stirrer commands, the userprovides the synthesis instructions in a text format similar to a written protocol. The graphicalrepresentation of the synthesis platform, which contains all necessary information about thefluidic connectivity, is represented by using GraphML. The written scheme from a publishedprocedure or lab book entry is translated into a series of ChASM commands. Both the ChASMand the GraphML files are passed to the Chempiler. The Chempiler command dispatcher thenuses those two pieces of information to control the physical labware through the respectivedevice drivers, effectively executing the synthesis.

RESEARCH | RESEARCH ARTICLEon F

ebruary 16, 2020

http://science.sciencemag.org/

Dow

nloaded from

amodified separating funnel and computer visionto directly replicate the way a human chemistwould perform liquid-liquid extractions. How-ever, we found that solutions using either acolored floater (17) or direct recognition of thephase boundary (18) worked well in a test envi-ronment but were otherwise unreliable. In realsyntheses, imperfections such as poor phase sepa-ration, strongly colored or cloudy solutions, orunusual extraction solvents often lead to com-plete failure of the image recognition algorithms.We therefore abandoned the computer vision fora sensor-based approach. Initially, we investigatedoptical and capacitive sensors because they donot require direct contact with the medium. Un-fortunately, those sensors also performed poorlyin some cases, so ultimately, we built a conduc-tivity sensor from two pieces of stainless-steeltubing inserted into the flow path (see SM). This

sensor reliably detected phase boundaries in alltest cases and enabled us to perform separationsin a robust fashion. The sensor was connectedto a custom-made separating funnel with a B45ground glass joint at the top and two B14 sidearms (fig. S36). Instead of a stopcock, it had aglass ¼-28 UNF (Unified fine) male threadedconnector (where ¼ indicates a major diameterof 0.25 inches, or 6.35 mm, and 28 indicates apitch of 28 threads per inch) fitted to the bottom.An Arduino Due was used to read the sensor viaa simple voltage divider circuit. The top inlet tubewas suspended by a ground glass joint–to–GL18thread adapter with a GL18 screw cap and insert.To facilitate efficient extractions through thor-ough mixing, a computer-controlled overheadstirrer was fitted above the separator.To perform liquid-liquid extractions, the mix-

ture was pumped into the separator through the

top inlet, or in the case of a wash, the washingsolvent was added through either the top or bot-tom port, depending on whether it constitutedthe top or bottom layer of the biphasic mixture.The two layerswere then stirred vigorously withthe overhead stirrer, and this step was followedby a period of settling under very slow (50 rpm)stirring. Next, the bulk of the lower phase wasusually transferred to the target vessel to speed upthe process. The volumemoved was determinedempirically for every separation. The actual sepa-ration commenced with withdrawing the deadvolume of the sensor and tubing from the bot-tom port. The volume removed was dispensedinto the lower-phase target vessel. Then a sensorreading was taken and compared against areference value. If the reading was lower thanthe reference, the lower phasewas assumed to beorganic; otherwise, the lower phasewas assumedto be aqueous. Onemilliliter was then transferredto the lower-phase target. Another sensor readingwas taken and compared against a predefinedthreshold value. This threshold depended onwhether the lower phasewas aqueous or organic.Either way, if the sensor reading was outside thethreshold, it was concluded that a phase changehad been detected. If not, another milliliter waswithdrawn and the cycle continued until a phasechange was detected. Then the dead volume ofthe sensor and tubing was transferred to thelower-phase target vessel. If the upper-phasetarget was specified as the separator, the pro-cess was concluded. Otherwise, the upper phasewas withdrawn as well and transferred to thetarget vessel.For the solvent evaporation, a computer-

controllable rotary evaporator was modifiedby routing a piece of PTFE tubing through thevapor duct into the evaporation flask to pumpproduct into and out of the flask (fig. S37). Thereceiver flask was fitted with a glass ¼-28 UNFmale threaded connector and a PTFE tube,allowing it to be emptied in situ. One complica-tion was that after distillation at reduced pres-sure, upon venting, oily products were forcedback into the tube reaching into the evapora-tion flask. This problem was solved by affixing amagnetic stirrer bar to the tube by using PTFEshrink wrap (fig. S38). A strongmagnet was thenpositioned in such a way that it would attract thetube and lift it out of the product, allowing thesystem to be vented. When the flask was loweredinto the heating bath, the tube would be releasedand drop, thereby allowing product to be with-drawn. Solvent evaporation startedwith pumpingthe solution to be evaporated into the distillationflask of the rotary evaporator. A cartridge filledwith molecular sieves could be switched into theflow path by two six-way selection valves, allow-ing for the solution to be dried before evapora-tion. The flask was then lowered into the heatingbath, and the rotationwas turnedon. The vacuumpump was started, lowering the pressure to900 mbar to degas the solution and avoid ex-cessive foaming later on. The heating bath andthe recirculation chiller servicing the condenserwere switched on, the target temperatures were

Steiner et al., Science 363, eaav2211 (2019) 11 January 2019 4 of 8

Syringe pump

Six-wayvalve

W To waste

1-12 Reagent positions

W WW

1

2 3

4 W

5 6

7 W

8

W

12

11

109

Backbone

Rotary evaporatorFilterReactor Separator

Rotaryevaporator

Backbone

ReactorReactor FilterFilter SeparatorSeparatorRotaryRotary

evaporatorevaporator

BackboneBackbone

A

B

Fig. 3. Physical implementation of the synthesis platform. (A) Schematic representation of theChemputer, highlighting modules used for four commonly encountered unit operations. The linesrepresent fluidic connections. (B) Photograph of one Chemputer setup used in this work.The variousmodules are highlighted in correspondence to the schematic.

RESEARCH | RESEARCH ARTICLEon F

ebruary 16, 2020

http://science.sciencemag.org/

Dow

nloaded from

set, and execution of the script was suspendeduntil the target temperatures were reached. Thevacuum set point was then changed to the tar-get distillation pressure, and the vacuum pumpspeed was adjusted according to the solvent, toavoid bumping. Then the execution of script wassuspended for a user-defined amount of timeto allow the main distillation to finish. After thedistillation was complete, the flask was lifted,which caused the inlet tube to be attracted by themagnet (fig. S38), lifting it out of the remainingsolution. The vacuum pump was subsequentlystopped, and the vacuum was vented. A user-defined amount of distillate was removed fromthe distillate flask and discarded. The param-eters (pressure, timings, and volumes) werealways chosen empirically, either through experi-mental trial and error or from prior experienceswith the system.At this point, we found that proceeding direct-

ly to drying the product under maximal vacuumwould often lead to a few milliliters of residualsolvent distilling over, which decreased the dry-ing efficiency. Thus, the flask was lowered backinto the bath and the vacuum pump was set tomaximum power for 2 min, drawing out any re-sidual solvent. The sequence of raising the flask,venting the vacuum, and emptying the distillateflask was then repeated. Next, the flask was low-ered once again, the vacuum pump was set tomaximum power and started, and the cooling ofthe condenser was switched off. The product wasthen dried for a user-defined amount of time.After the dryingwas complete, the flaskwas liftedoncemore, the vacuumwas vented, and the rota-tion and heating bath were switched off.The filtration unit consisted of a custom-made,

jacketed, sintered glass Büchner funnel (madein-house; see SM) fitted with a B29 ground glassjoint at the top, two B14 side arms, and a glass¼-28UNFmale threaded connector at the bottom.The top inlet tube was suspended by a groundglass joint–to–GL18 thread adapter with a GL18screw cap and insert, whereas the bottom outlettubewas connected to the threaded connectorwitha straight union piece. Stirringwas accomplishedby a computer-controllable overhead stirrer.To allow efficient drying of the precipitate, the

bottom outlet of the filter was connected to thecentral inlet of a six-way valve. One outlet of thatvalve was then connected to the backbone, andanother outlet was connected to the laboratoryvacuum system via aWoulff bottle. This allowedthe user to switch the filter bottom between thebackbone (for liquid addition or withdrawal) andvacuum (for drying). The whole platform couldbe cleaned automatically by pumping suitablewashing solvents into themodules. This cleaningcycle would return the system to its initial state,ready for the next reaction stage. Consequently,preparing the system for another synthesis wassimply amatter of swapping the reagent bottles.

Proof-of-principle automated synthesesof three drugs

To highlight the power of this approach, wechose three targets: diphenhydramine hydro-

chloride (6) (Fig. 4), rufinamide (10) (Fig. 5), andsildenafil (17) (Fig. 6). The process of digitizing asynthesis always starts with a traditional, writtenscheme such as an experimental procedure froma publication or a lab book entry. We chose threepublished syntheses, all replicated manually toestablish benchmark yields and purities for com-parison with the automated runs.Diphenhydramine hydrochloride is an etha-

nolamine derivative used as an antihistamineand amild sleep aid. It ismarketed asNytol in theUnited Kingdom and as Benadryl in the UnitedStates. The synthesis is a four-step sequencestarting with a Grignard reaction. Rufinamideis a triazole derivative used as an anticonvulsantto treat various seizure disorders, and its syn-thesis is a relatively simple process to automate.Sildenafil is prescribed to treat erectile dysfunc-tion and is best known under the brand nameVIAGRA; its industrial synthesis route (19) fea-tures a chlorosulfonation with highly aggres-sive chlorosulfonic acid and thionyl chloride. Wereasoned that successful automated handlingof those aggressive reagents would demonstratethe versatility and robustness of the system, aswell as highlighting the safety benefit arisingfrom automating dangerous procedures.

After reproducing the synthesis of diphen-hydramine hydrochloride (6) (20) manually, wemade some small modifications and started theprocess on the platform. The synthesis commencedwith Grignard reagent formation, for which thereactor was manually charged with dried mag-nesium grit. All the other required reagents andsolvents were loaded into 100- or 250-ml stan-dard GL45 bottles, and all nonaqueous storagebottles were purgedwith argon and stored underpositive pressure. All the operations describedwere performed by the Chemputer under fullChempiler control, and the programused, aswellas a description of the process in prose, can befound in the SM. The automated synthetic pro-cedure was started by automatically priming thetubes to the chemical reservoirs and then auto-cleaning the backbone with water, isopropanol,and dry diethyl ether. The system then continuedautonomously through the whole synthesis ofdiphenhydramine hydrochloridewithout humanintervention as follows. Diethyl ether and a smallportion of bromobenzene were added to themag-nesium, and themixture was heated under refluxto initiate the Grignard reagent formation. Aftercooling, the remaining bromobenzene was addedat a rate of 1 ml/min, and the mixture was again

Steiner et al., Science 363, eaav2211 (2019) 11 January 2019 5 of 8

Mg

Et2O, reflux

CH3COBr

toluene, reflux

OH BrBr

H

O

+

1 2 3 4

toluene, reflux

HON

Et2O, 25°C

HCl·Et2O

ON

ON

·HCl

5 6

Liquid/LiquidSeparator

ReactorModule

FiltrationModule

EvaporationModule

TolueneCH

3COBr

Et2O

Mg

Bromo-benzene

Benz-aldehyde

Waste Waste

H2OHCl

TolueneDMAE

NaOH HCl

Waste

NaOHEt

2O

HCl Et2O 2-propanol

Waste Waste Waste

A

B

Fig. 4. Synthesis of diphenhydramine hydrochloride (6). (A) Modified synthetic route todiphenhydramine hydrochloride. Et, ethyl. (B) Sequence of unit operations required for the synthesis.The dotted boxes denote the four stages of the synthesis. DMAE, dimethylaminoethanol.

RESEARCH | RESEARCH ARTICLEon F

ebruary 16, 2020

http://science.sciencemag.org/

Dow

nloaded from

heated to reflux. Using syringe pumps for movingmaterial allowed us to precisely control additionrates and thus increase the reproducibility of syn-thetic protocols even further. Subsequently, a solu-tion of benzaldehyde in diethyl ether was addedat a rate of 1 ml/min, and themixture was heldat reflux for another 5 hours. As the platformpresented herein is largely a proof of concept,no process analytical technology (PAT) has beenimplemented yet, so all reaction times were de-termined beforehand and hard-coded into theChASM script. However, the modular architec-ture of the platform and control software shouldmake adding PAT to future iterations of the plat-form straightforward. After quenching of the re-actionwith dilute HCl, the layers were separated,and the organic layer was washed with waterand concentrated in vacuo as described earlier.The system then cleaned itself and directly pro-ceeded with the bromination. To that end, thereactor was charged with acetyl bromide, thecrude diphenylmethanol (3) was transferredfrom the rotary evaporator flask to the reactorwith three portions of toluene, and the mixturewas heated to reflux, all without human inter-vention. After a predetermined reaction time,the mixture was transferred to the rotary evapo-rator and evaporated to dryness. The subsequentWilliamson ether synthesis was initiated in asimilar fashion, and after a predetermined time,the reaction was quenched with aqueous sodiumhydroxide. The system then automatically con-ducted an aqueousworkup and concentrated theproduct in vacuo. Once again, the system cleaneditself, charged the jacketed filtrationmodulewithhydrochloric acid, and transferred the crude freebase 5 to the filter with three portions of diethylether. To ensure smooth precipitation, the mix-ture was stirred vigorously and the free base

solution was added very slowly. After the addi-tionwas completed, the off-white precipitate wascollected by automatic filtration and recrystal-lized from isopropanol by using the heating andcooling capabilities of the jacketed filter. Dryingat 60°C in a stream of argon for 1 hour yieldedpure diphenhydramine hydrochloride, giving anisolated yield of 58% over four steps, or 87% perstep on average. Although this is slightly less thanthe 68% overall achievedmanually, the averageyield per step (87% for automated versus 91%for manual) is comparable in our view. The plat-form performed the synthesis fully automaticallyin 77 hours (see movie S1), whereas the manualsynthesis took 4 work days.Next, we conducted an automated synthesis of

the antiseizure drug rufinamide (10), a triazolederivative commonly prepared via a click reactionbetween the corresponding azide 8 and methylpropiolate (Fig. 5) (21). The synthesis began withan azide formation, for which the reactor wascharged manually with 2,6-difluorobenzyl bro-mide (7); the remaining reagents were providedin bottles as described above. From this point on,unless explicitly stated otherwise, all describedoperations were performed by the Chemputerunder full Chempiler control. An aqueous solu-tion of sodium azide was added to the reactor toprepare the organic azide for the triazole clickwith methyl propiolate, which was performedinside the jacketed filtration module. Subse-quent saponificationwith aqueous ammonia ledto precipitation of the target compound. Filtra-tion followed by three aqueous washes yieldedpure rufinamide at 46% isolated yield, whichwasslightly better than the manual synthesis (38%).The automated synthesis took 38 hours. To dem-onstrate the power of the Chempiler softwareand the interoperability of the code, we thenwent

on to run the same ChASM file on a “full-scale”platform equippedwith slightly different hardware,which was connected in an entirely different way.The platform produced pure rufinamide in 44%yield without any problems or changes to thecode (movie S2).In the next synthesis, we prepared sildenafil,

better knownunder the brandnameVIAGRA (17),as shown in Fig. 6 (13). For this synthesis, we fittedthe reactor with a heating block connected to therecirculation chiller, allowing us to cool as well asheat. From this point on, unless explicitlystated otherwise, all described operations wereperformedby theChemputer under full Chempilercontrol. The reactor was cooled to 15°C and auto-matically chargedwith chlorosulfonic acid, thionylchloride, and molten ethoxybenzoic acid. Chlo-rosulfonic acid is corrosive, so when writing theChASM script we took great care to minimizecontact times and enacted a strict cleaning re-gime. Chlorosulfonic acid and thionyl chloridealso react violently with trace amounts of water,producing large volumes of gas. Therefore, thebackbone was automatically flushed with drydiethyl ether and dried with a small amount ofthionyl chloride before the reactor was charged.After a predetermined reaction time, the filtra-tionmodulewas chargedwithwater and cooledto 0°C. The reaction mixture was then slowlydripped into the water, quenching the excessthionyl chloride and chlorosulfonic acid andprecipitating the product (12), which was col-lected by automated filtration. The subsequentsulfonamide formation withN-methylpiperazine(13) in water was performed in the filtrationmodule as well. Unfortunately, the sulfonamide(14) did not crystallize spontaneously, so a slurryof a small amount of product in water was addedto seed the crystallization.The industrial process for sildenafil uses N,

N′-carbonyldiimidazole for the amide couplingin the next step. However, this reaction did notwork in our hands, either manually or in auto-mation, and thus we decided to use the acidchloride instead. The carboxylic acid (14) wasthoroughly dried by flowing argon through thefilter cake while at the same time heating thefiltrationmodule to 60°C, after which acid chlo-ride formed with thionyl chloride in dichloro-methane. The reactor module was charged witha solution of 4-amino-1-methyl-3-n-propyl-1H-pyrazole-5-carboxamide (15) in dichloromethaneand triethylamine and cooled to 10°C. The crudeacid chloride solution was then pumped fromthe filtermodule to the reactor, the reactionwasquenchedwith water, the layers were separated,and the organic layer was dried over activatedmolecular sieves and evaporated to dryness,yielding amide 16 as an off-white solid. For thesubsequent cyclization, the crude amide wastransferred back to the reactor by dissolvingit with a solution of potassium tert-butoxidein tert-butanol, and the mixture was heated atreflux for 8 hours. After the mixture was cooledto 10°C, the reaction was quenched with water,and the solution was transferred to the filtermodule. To induce precipitation of the sildenafil,

Steiner et al., Science 363, eaav2211 (2019) 11 January 2019 6 of 8

7 8 9

H2O

CO2MeNaN3

H2O, MeOH

F

F

Br

F

F

N3

F

F

NN

N

CO2Me

10

NH3

H2O

F

F

NN

N

ONH2

ReactorModule

FiltrationModule

H2O

NaN3

Difluorobenzyl-bromide

Methylpropiolate NH

3

Waste

A

B

Fig. 5. Synthesis of rufinamide (10). (A) Synthetic route to rufinamide. Me, methyl. (B) Sequenceof unit operations required for the synthesis.

RESEARCH | RESEARCH ARTICLEon F

ebruary 16, 2020

http://science.sciencemag.org/

Dow

nloaded from

themixturewas neutralizedwith aqueous hydro-chloric acid. After filtration, the solidwaswashedwith water and dried under a stream of argonat 50°C, yielding sildenafil at 44% isolated yieldover 102 hours (movie S3).

Outlook

The complete automation of all of synthesis is anambitious objective, but this work makes a starttoward that goal, as the Chemputer architecturepresents a general abstraction of the process thatworks with traditional bench-scale techniques.The versatile programming language, the use oftraditional and inexpensive labware (the totalcost of the parts for the robotic modules, includ-ing nonstandard glassware, is less than $10,000per system) (22), and the payoff in reproducibilityafter validation of the process mean that adop-tion could be straightforward. Initially, the syn-thesis of compounds will be validated reaction byreaction, but we imagine that eventually it will bepossible to go straight from a reaction database toa chemical code that can run the platforms.

Materials and methods summary

The manual and automated syntheses of thethree target molecules, as well as the platformand the control software, are described in detailin the SM, and this information has been depo-sited in Zenodo (23), along with the ChASM andGraphML code. Videos of the automated synthe-

ses are available as movies S1 to S3. We willcontinue to update the ChASM code for the syn-theses, and updates of this will be available todownload (24). A brief summary of the synthesesis provided below.

Synthesis of diphenhydraminehydrochloride

The reactor module was charged manually withmagnesium grit and dried by heating to 150°Cunder a stream of argon for 15 min. After cool-ing to room temperature (~25°C), diethyl etherand 2 ml of bromobenzene were added to themagnesium, and themixturewas heated to refluxfor 20 min. After cooling below 25°C, 8.65 ml ofbromobenzene was added at a rate of 1 ml/min,and the mixture was again heated to reflux for20min. Subsequently, a solution of benzaldehydein diethyl ether was added at 1 ml/min, and themixture was held at reflux for 5 hours. Afterquenching of the reaction with dilute HCl, thelayers were separated, and the organic layer waswashed with water and evaporated to dryness,yielding crude diphenylmethanol. After auto-matic cleaning of the system, the reactor wascharged with acetyl bromide, and the crude di-phenylmethanol was transferred from the rotaryevaporator to the reactor with three portionsof toluene. Themixture was heated to reflux for4 hours and subsequently evaporated to dryness,yielding crude bromodiphenylmethane. The sys-

tem was automatically cleaned once more, the re-actorwas chargedwith 2-(dimethylamino)ethanoland 10 ml of toluene, and the bromodiphenyl-methanewas transferred to the reactor with threeportions of toluene. The mixture was heated toreflux for 20 hours. After cooling below 30°C,the reaction was quenched with aqueous sodiumhydroxide. The layers were separated, and theorganic layer was extracted with 2 M aqueoushydrochloric acid three times. Equimolar aqueoussodium hydroxide was added to the combinedaqueous layers, and the mixture was extractedwith diethyl ether three times. The combinedetheric layerswere evaporated to dryness, yieldingcrude diphenhydramine free base. The jacketedfilter module was then charged with etheric hy-drochloric acid, and the crude free base was slow-ly transferred to the filter with three portions ofdiethyl ether. The precipitate formedwas collectedby filtration, dried under a stream of argon, andrecrystallized from isopropanol, yielding purediphenhydramine hydrochloride as white crys-talline powder.

Synthesis of rufinamide

The reactor was charged manually with 2,6-difluorobenzyl bromide. An aqueous solutionof sodium azide was added to the reactor, andthe mixture was heated to 75°C for 12 hoursand subsequently transferred to the jacketedfiltermodule. Neatmethyl propiolatewas added,and the mixture was heated to 65°C for 4 hours.An aqueous solution of ammonia was subse-quently added, and the mixture was held at 75°Cfor an additional 12 hours, precipitating thetarget compound. Filtration followed by threeaqueous washes yielded pure rufinamide aswhite crystalline powder.

Synthesis of sildenafil

The reactor was automatically charged withchlorosulfonic acid, thionyl chloride, andmoltenethoxybenzoic acid. The mixture was stirred at15°C for 30min and then at room temperature foran additional 12 hours. Subsequently, the filtra-tionmodule was charged with water and cooledto 0°C. The reactionmixture was slowly drippedinto the water, quenching the excess thionylchloride and chlorosulfonic acid and precipitat-ing the product. The supernatant solution was re-moved, and the precipitatewaswashedwith coldwater, yielding 5-chlorosulfonyl-2-ethoxybenzoicacid as white powder. The wet solid remainingin the filtermodule was subsequently slurried incold water, and N-methylpiperazine was addedslowly. After 5 min, crystallization was initiatedby adding a slurry of a small amount of product.The solid was collected by filtration, washedwith cold water, and dried under a stream ofargon at 50°C, yielding 2-ethoxy-5-(4-methyl-1-piperazinesulfonyl)-benzoic acid as a whitepowder. The dry carboxylic acid was slurried indichloromethane and cooled to 5°C. Thionylchloride and a catalytic amount of dimethyl-formamide were added, and the mixture wasstirred for 5 hours at 25°C. Subsequently, thereactor module was charged with a solution of

Steiner et al., Science 363, eaav2211 (2019) 11 January 2019 7 of 8

11

OEtCOOH

12

OEtCOOH

SO2Cl

14

OEtCOOH

O2S NNMe

13HN

NMe

H2O

HSO3Cl

SOCl2

16

OEt

NH

O

O2S N

NMe

NMeN

CONH2

nPr

tBuOK

tBuOH

15

1) SOCl2

CH2Cl2

MeN

NH2N

O

H2NnPr

2)

O2S N

NMe

N

HNEtO

O

N

MeN

nPr

17

Liquid/LiquidSeparator

ReactorModule

FiltrationModule

EvaporationModule

15

A

B

Fig. 6. Synthesis of sildenafil (17). (A) Synthetic route to sildenafil. tBu, tert-butyl;nPr, normal-propyl. (B) Sequence of unit operations required for the synthesis.

RESEARCH | RESEARCH ARTICLEon F

ebruary 16, 2020

http://science.sciencemag.org/

Dow

nloaded from

4-amino-1-methyl-3-n-propyl-1H-pyrazole-5-carboxamide in dichloromethane and triethyl-amine and cooled to 10°C. The crude acid chloridesolutionwas pumped from the filtermodule to thereactor, and themixture was stirred for 16 hoursat 25°C. After the reaction was quenched withwater, the layers were separated, and the organiclayer was dried over activated molecular sievesand evaporated to dryness, yielding 4-[2-ethoxy-5-(4-methyl-1-piperazinylsulfonyl)-benzamido]-1-methyl-3-propyl-1H-pyrazole-5-carboxamide asan off-white solid. For the subsequent cycliza-tion, the crude amide was transferred back to thereactor by dissolving it with a solution of po-tassium tert-butoxide in tert-butanol, and themixture was heated at reflux for 8 hours. Aftercooling to 10°C, the reactionwas quenchedwithwater, and the solution was transferred to thefilter module. To induce precipitation of thesildenafil, the mixture was neutralized withaqueous hydrochloric acid. After filtration, thesolid was washed with water and dried under astream of argon at 50°C, yielding the sildenafilas white crystalline powder.

REFERENCES AND NOTES

1. J. Li et al., Synthesis of many different types of organicsmall molecules using one automated process. Science 347,1221–1226 (2015). doi: 10.1126/science.aaa5414; pmid: 25766227

2. P. J. Kitson et al., Digitization of multistep organic synthesisin reactionware for on-demand pharmaceuticals. Science 359,314–319 (2018). doi: 10.1126/science.aao3466; pmid: 29348235

3. C. Rougeot, H. Situ, B. H. Cao, V. Vlachos, J. E. Hein,Automated reaction progress monitoring of heterogeneousreactions: Crystallization-induced stereoselectivity inamine-catalyzed aldol reactions. React. Chem. Eng. 2, 226–231(2017). doi: 10.1039/C6RE00211K

4. F. Häse, L. M. Roch, C. Kreisbeck, A. Aspuru-Guzik, Phoenics:A Bayesian optimizer for chemistry. ACS Cent. Sci. 4, 1134–1145(2018). doi: 10.1021/acscentsci.8b00307; pmid: 30276246

5. A. Adamo et al., On-demand continuous-flow production ofpharmaceuticals in a compact, reconfigurable system.Science 352, 61–67 (2016). doi: 10.1126/science.aaf1337;pmid: 27034366

6. A. G. Godfrey, T. Masquelin, H. Hemmerle, A remote-controlledadaptive medchem lab: An innovative approach to enable drugdiscovery in the 21st century. Drug Discov. Today 18, 795–802(2013). doi: 10.1016/j.drudis.2013.03.001; pmid: 23523957

7. R. B. Merrifield, Automated synthesis of peptides. Science 150,178–185 (1965). doi: 10.1126/science.150.3693.178;pmid: 5319951

8. G. Alvarado-Urbina et al., Automated synthesis of genefragments. Science 214, 270–274 (1981). doi: 10.1126/science.6169150; pmid: 6169150

9. M. B. Plutschack, B. Pieber, K. Gilmore, P. H. Seeberger, Thehitchhiker’s guide to flow chemistry. Chem. Rev. 117,11796–11893 (2017). doi: 10.1021/acs.chemrev.7b00183;pmid: 28570059

10. O. J. Plante, E. R. Palmacci, P. H. Seeberger, Automatedsolid-phase synthesis of oligosaccharides. Science 291,1523–1527 (2001). doi: 10.1126/science.1057324; pmid: 11222853

11. P. D. Karp, Pathway databases: A case study in computationalsymbolic theories. Science 293, 2040–2044 (2001).doi: 10.1126/science.1064621; pmid: 11557880

12. M. Baker, 1,500 scientists lift the lid on reproducibility.Nature 533, 452–454 (2016). doi: 10.1038/533452a;pmid: 27225100

13. GraphML, http://graphml.graphdrawing.org/ [accessed19 October 2018].

14. D. G. Brown, J. Boström, Analysis of past and presentsynthetic methodologies on medicinal chemistry: Where haveall the new reactions gone? J. Med. Chem. 59, 4443–4458(2016). doi: 10.1021/acs.jmedchem.5b01409; pmid: 26571338

15. A. C. Bédard et al., Reconfigurable system for automatedoptimization of diverse chemical reactions. Science 361,1220–1225 (2018). doi: 10.1126/science.aat0650;pmid: 30237351

16. A. Adamo, P. L. Heider, N. Weeranoppanant, K. F. Jensen,Membrane-based, liquid-liquid separator with integratedpressure control. Ind. Eng. Chem. Res. 52, 10802–10808(2013). doi: 10.1021/ie401180t

17. M. O’Brien, P. Koos, D. L. Browne, S. V. Ley, A prototypecontinuous-flow liquid-liquid extraction system usingopen-source technology. Org. Biomol. Chem. 10, 7031–7036(2012). doi: 10.1039/c2ob25912e; pmid: 22858870

18. S. Eppel, T. Kachman, Computer vision–based recognition ofliquid surfaces and phase boundaries in transparent vessels,with emphasis on chemistry applications. arXiv:1404.7174[cs.CV] (6 November 2014).

19. D. J. Dale et al., The chemical development of the commercialroute to sildenafil: A case history. Org. Process Res. Dev. 4,17–22 (2000). doi: 10.1021/op9900683

20. A. Ahmadi, M. Khalili, R. Hajikhani, N. Safari, B. Nahri-Niknafs,Anti-inflammatory effects of two new methyl and morpholine

derivatives of diphenhydramine on rats. Med. Chem. Res. 21,3532–3540 (2011). doi: 10.1007/s00044-011-9891-y

21. R. D. Padmaja, K. Chanda, A short review on synthetic advancestoward the synthesis of rufinamide, an antiepileptic drug. Org.Process Res. Dev. 22, 457–466 (2018). doi: 10.1021/acs.oprd.7b00373

22. The total cost of the parts for robotic modules, includingnonstandard glassware, is less than $10,000 per system, butthis rises to $30,000 if the stirrers, evaporator, and chillersystem are included, although these components mightreasonably be expected to be found in most well-equippedsynthesis laboratories.

23. G. Keenan, croningp/ChemputerSoftware: Chemputer firstrelease, Zenodo (2018). doi: 10.5281/zenodo.1481731

24. Chemify; www.chemify.org.

ACKNOWLEDGMENTS

We thank M. Craven for help with the XDL development andH. Mehr, S. Zalesskiy, M. Symes, and R. Hartley for comments on themanuscript. Funding: We gratefully acknowledge financial supportfrom the EPSRC (grants EP/H024107/1, EP/J015156/1, EP/K021966/1, EP/L015668/1, and EP/L023652/1) and the ERC(project 670467 SMART-POM). Author contributions: L.C.conceived the initial concept design approach and coordinated theteam. P.J.K., G.A.-C., G.K., and L.C. developed the abstraction.L.C., T.H., and S.G. developed the fluidic backbone and initialversions of the pumps and valves, including the electronics andfirmware. T.H. outlined the initial XDL with L.C. S.S. built thefinal versions of the rig, developed ChASM, and conducted thesyntheses with help from J.W., A.A., and J.M.G. D.A. helped withthe sildenafil automation. G.K. and G.A.-C. developed the softwaresystem with S.S. L.C. and S.S. wrote the manuscript. Competinginterests: L.C. is the founder and director of DeepMatter GroupPLC (deepmatter.io), which aims to digitize chemistry. Dataand materials availability: All the software to run the Chemputerto synthesize the compounds described herein has been depositedin Zenodo (23).

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/363/6423/eaav2211/suppl/DC1Materials and MethodsSupplementary TextFigs. S1 to S58Movies S1 to S3Data S1 and S2

30 August 2018; accepted 12 November 2018Published online 29 November 201810.1126/science.aav2211

Steiner et al., Science 363, eaav2211 (2019) 11 January 2019 8 of 8

RESEARCH | RESEARCH ARTICLEon F

ebruary 16, 2020

http://science.sciencemag.org/

Dow

nloaded from

Organic synthesis in a modular robotic system driven by a chemical programming language

Aragon-Camarasa, Philip J. Kitson, Davide Angelone and Leroy CroninSebastian Steiner, Jakob Wolf, Stefan Glatzel, Anna Andreou, Jaroslaw M. Granda, Graham Keenan, Trevor Hinkley, Gerardo

originally published online November 29, 2018DOI: 10.1126/science.aav2211 (6423), eaav2211.363Science 

, this issue p. eaav2211; see also p. 122Sciencepharmaceuticals that proceeded comparably to manual synthesis.compatibility with extant literature. The authors showcase the system with short syntheses of three common conventional equipment such as round-bottom flasks, separatory funnels, and a rotary evaporator to maximize itscompounds on the basis of standardized methods descriptions (see the Perspective by Milo). The platform comprises

developed an autonomous compiler and robotic laboratory platform to synthesize organicet al.and for humans. Steiner The chemistry literature contains more than a century's worth of instructions for making molecules, all written by

Clear directions for a robotic platform

ARTICLE TOOLS http://science.sciencemag.org/content/363/6423/eaav2211

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2018/11/28/science.aav2211.DC1

CONTENTRELATED http://science.sciencemag.org/content/sci/363/6423/122.full

REFERENCES

http://science.sciencemag.org/content/363/6423/eaav2211#BIBLThis article cites 19 articles, 8 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science. No claim to original U.S. Government WorksCopyright © 2019 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on February 16, 2020

http://science.sciencem

ag.org/D

ownloaded from