CoChat

Self-Driving Laboratories for Chemistry and Materials Science

Open access

Published version of a preprint

The DOI matches its Crossref record; 732 of the 732 references it cites were found in a registry or an index.

Abstract

Self-driving laboratories (SDLs) promise an accelerated application of the scientific method. Through the automation of experimental workflows, along with autonomous experimental planning, SDLs hold the potential to greatly accelerate research in chemistry and materials discovery. This review provides an in-depth analysis of the state-of-the-art in SDL technology, its applications across various scientific disciplines, and the potential implications for research and industry. This review additionally provides an overview of the enabling technologies for SDLs, including their hardware, software, and integration with laboratory infrastructure. Most importantly, this review explores the diverse range of scientific domains where SDLs have made significant contributions, from drug discovery and materials science to genomics and chemistry. We provide a comprehensive review of existing real-world examples of SDLs, their different levels of automation, and the challenges and limitations associated with each domain.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1021/acs.chemrev.4c00055

Cites

732 of 732 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Schematic for the autonomy levels of SDLs based on the category of hardware and software autonomy achieved.
Figure 1 Schematic for the autonomy levels of SDLs based on the category of hardware and software autonomy achieved.CC BY-NC-ND · PubMed Central
Figure 2 Examples of types of automated hardware. (a)-(c) Categories of automated hardware for SDLs and (d)-(e) supporting software for automation. Bottom half show examples of each category: (a) OT-2 platform manufactured by OpenTrons;41(b) robotic arm for chemical operations in Cooper’s autonomous laboratory, adapted with permission from reference.11 Copyright 2020, Springer Nature; (c) 3D schematics of Sidekick, a low cost liquid dispensing platform developed by Keesey et al., adapted with permission from reference.42 Copyright 2022, Elsevier; (d) computer vision framework for laboratory glassware developed by Eppel et al.;12 and (e) solid weighting simulation software developed by Kadokawa et al., adapted with permission from reference.43 Copyright 2023, Institute of Electrical and Electronics Engineers (IEEE).
Figure 2 Examples of types of automated hardware. (a)-(c) Categories of automated hardware for SDLs and (d)-(e) supporting software for automation. Bottom half show examples of each category: (a) OT-2 platform manufactured by OpenTrons;41(b) robotic arm for chemical operations in Cooper’s autonomous laboratory, adapted with permission from reference.11 Copyright 2020, Springer Nature; (c) 3D schematics of Sidekick, a low cost liquid dispensing platform developed by Keesey et al., adapted with permission from reference.42 Copyright 2022, Elsevier; (d) computer vision framework for laboratory glassware developed by Eppel et al.;12 and (e) solid weighting simulation software developed by Kadokawa et al., adapted with permission from reference.43 Copyright 2023, Institute of Electrical and Electronics Engineers (IEEE).CC BY-NC-ND · PubMed Central
Figure 3 Summary of the software components of SDLs, and how they interface with the hardware, and with each other.
Figure 3 Summary of the software components of SDLs, and how they interface with the hardware, and with each other.CC BY-NC-ND · PubMed Central
Figure 4 Depiction of the different encoding techniques (left) and ML/DL models (right) used in SDLs. They are commonly combined to obtain a deeper knowledge of the studied system and acceleration of chemical exploration through experimental proposals.
Figure 4 Depiction of the different encoding techniques (left) and ML/DL models (right) used in SDLs. They are commonly combined to obtain a deeper knowledge of the studied system and acceleration of chemical exploration through experimental proposals.CC BY-NC-ND · PubMed Central
Figure 5 Autonomous optimization of chromatographic separation by varying the eluent composition. Top: Schematic visualization of the experimental setup used by Berridge, and structures of four analytes in a sample mixture. Bottom: Visualization of the simplex optimization performance upon variation of eluent composition and flow rate (left). Development of the chromatographic response function (CRF), the objective, throughout the course of the optimization campaign (middle); the optimized chromatogram (right). Figure adapted with permission from reference.257 Copyright 1982, Elsevier.
Figure 5 Autonomous optimization of chromatographic separation by varying the eluent composition. Top: Schematic visualization of the experimental setup used by Berridge, and structures of four analytes in a sample mixture. Bottom: Visualization of the simplex optimization performance upon variation of eluent composition and flow rate (left). Development of the chromatographic response function (CRF), the objective, throughout the course of the optimization campaign (middle); the optimized chromatogram (right). Figure adapted with permission from reference.257 Copyright 1982, Elsevier.CC BY-NC-ND · PubMed Central
Figure 6 Purity optimization results in multi-step reaction-extraction process. Four parameters are varied: the temperature, solvent ratio (VR), the residence time, and the inlet pH. The size of the dots corresponds to the temperature, and the color represents the purity of N-benzyl-α-methylbenzylamine after the reaction. The star is the optimal purity obtained at experiment 53. Figure reproduced with permission from Clayton et al.362 Copyright 2020, Springer Nature.
Figure 6 Purity optimization results in multi-step reaction-extraction process. Four parameters are varied: the temperature, solvent ratio (VR), the residence time, and the inlet pH. The size of the dots corresponds to the temperature, and the color represents the purity of N-benzyl-α-methylbenzylamine after the reaction. The star is the optimal purity obtained at experiment 53. Figure reproduced with permission from Clayton et al.362 Copyright 2020, Springer Nature.CC BY-NC-ND · PubMed Central
Figure 7 General components of a closed-loop system for reaction condition optimization, consisting of an automated reactor system for automated reaction execution, a monitoring system for quantifying the reaction outcome, and an optimization algorithm for decision-making.
Figure 7 General components of a closed-loop system for reaction condition optimization, consisting of an automated reactor system for automated reaction execution, a monitoring system for quantifying the reaction outcome, and an optimization algorithm for decision-making.CC BY-NC-ND · PubMed Central
Figure 8 Example of an early automated robotic synthesizer centered around a Zymark robotic system. A: Robotic arm. B: Reactor station. C: Remote dispenser. D: Aliquot archive station. E: Workup station. F: Syringe and needle wash station. G: Turntable. H: Reporting integrator for analytical instrumentation. I: Reagent station. J: Robotic tool parking station. Figure reproduced with permission from Frisbee et al.391 Copyright 1984, American Chemical Society.
Figure 8 Example of an early automated robotic synthesizer centered around a Zymark robotic system. A: Robotic arm. B: Reactor station. C: Remote dispenser. D: Aliquot archive station. E: Workup station. F: Syringe and needle wash station. G: Turntable. H: Reporting integrator for analytical instrumentation. I: Reagent station. J: Robotic tool parking station. Figure reproduced with permission from Frisbee et al.391 Copyright 1984, American Chemical Society.CC BY-NC-ND · PubMed Central
Figure 9 Schematic of early example of autonomous reaction condition optimization in solution performed by Matsuda et al.258 Coupling reaction of carboxylic acids with N,N’-dicyclohexylcarbodiimide (DCC) and hydroxylamine, and detection of the resulting hydroxamic acid as its iron(III) complex through UV-Vis absorption spectroscopy.
Figure 9 Schematic of early example of autonomous reaction condition optimization in solution performed by Matsuda et al.258 Coupling reaction of carboxylic acids with N,N’-dicyclohexylcarbodiimide (DCC) and hydroxylamine, and detection of the resulting hydroxamic acid as its iron(III) complex through UV-Vis absorption spectroscopy.CC BY-NC-ND · PubMed Central
Figure 10 Closed-loop optimization of porphyrin synthesis conditions, as reported by Lindsey and co-workers. Optimization was performed under variation of the concentration of reactants, as well as the quantity of trifluoroacetic acid (TFA). The bottom row shows the response surface, as obtained by factorial design experiments (left), as well as the optimization trajectory of the simplex algorithm (right). Figure adapted with permission from reference.394 Copyright 1992, Elsevier. DDQ: Dichlorodihydroquinone.
Figure 10 Closed-loop optimization of porphyrin synthesis conditions, as reported by Lindsey and co-workers. Optimization was performed under variation of the concentration of reactants, as well as the quantity of trifluoroacetic acid (TFA). The bottom row shows the response surface, as obtained by factorial design experiments (left), as well as the optimization trajectory of the simplex algorithm (right). Figure adapted with permission from reference.394 Copyright 1992, Elsevier. DDQ: Dichlorodihydroquinone.CC BY-NC-ND · PubMed Central
Figure 11 Micro-reactor self-optimizing platform for a Knoevenagel condensation and the oxidation of benzyl alcohol towards benzaldehyde. (a) Overview of the system consisting of a control center to adjust residence time, temperature and concentrations to control the flow from the syringe pumps in the micro-reactor. Products are detected via HPLC, which returns the data to the control unit. (b) Image of the used micro-reactor. (c) Packaging scheme for the microreactor including fluidic connections in the top plate (1), a recessed plate (2) to house the microreactor and TE device, and baffled heat exchanger (3) for sufficient heat removal and additional temperature control. Figure adapted with permission from McMullen et al.268 Copyright 2010, American Chemical Society.
Figure 11 Micro-reactor self-optimizing platform for a Knoevenagel condensation and the oxidation of benzyl alcohol towards benzaldehyde. (a) Overview of the system consisting of a control center to adjust residence time, temperature and concentrations to control the flow from the syringe pumps in the micro-reactor. Products are detected via HPLC, which returns the data to the control unit. (b) Image of the used micro-reactor. (c) Packaging scheme for the microreactor including fluidic connections in the top plate (1), a recessed plate (2) to house the microreactor and TE device, and baffled heat exchanger (3) for sufficient heat removal and additional temperature control. Figure adapted with permission from McMullen et al.268 Copyright 2010, American Chemical Society.CC BY-NC-ND · PubMed Central
Figure 12 Self-optimization campaigns for the dehydration of ethanol and methylation of 1-pentanol in supercritical CO2 as reported by Poliakoff and co-workers. The optimizations varied the CO2 flow rate, temperature and pressure to optimize the yield of the products colored in brown. Bottom figures show the layout of the used reaction hardware (left) and the optimization campaign of the final reaction (right). Figure adapted with permission from Parrot et al.397 Copyright 2011, John Wiley and Sons.
Figure 12 Self-optimization campaigns for the dehydration of ethanol and methylation of 1-pentanol in supercritical CO2 as reported by Poliakoff and co-workers. The optimizations varied the CO2 flow rate, temperature and pressure to optimize the yield of the products colored in brown. Bottom figures show the layout of the used reaction hardware (left) and the optimization campaign of the final reaction (right). Figure adapted with permission from Parrot et al.397 Copyright 2011, John Wiley and Sons.CC BY-NC-ND · PubMed Central
Figure 13 Demonstration of a ReactIR system in the self-optimization of the Paal-Knorr synthesis of a pyrrole. In the bottom, a scheme of the utilized flow system is demonstrated along with results of the two-dimensional optimization campaign (left) and a sample IR spectrum (right). Figure adapted with permission from Moore and Jensen399 Copyright 2012, American Chemical Society.
Figure 13 Demonstration of a ReactIR system in the self-optimization of the Paal-Knorr synthesis of a pyrrole. In the bottom, a scheme of the utilized flow system is demonstrated along with results of the two-dimensional optimization campaign (left) and a sample IR spectrum (right). Figure adapted with permission from Moore and Jensen399 Copyright 2012, American Chemical Society.CC BY-NC-ND · PubMed Central
Figure 14 Demonstration of an NMR system in the self-optimization of the condensation between aniline and p-Fluorobenzaldehyde. (A) Scheme of the utilized flow system is demonstrated showing the input streams of the reagents and trifluoroacetic acid (TFA) as catalyst, leading to the NMR spectrometer for analytical measurement. (B) Sample NMR spectrum from the optimization campaign, indicating characteristic product and reactant peaks. (C) Results of the optimization campaign to optimize objective function J over multiple iterations. Figure adapted with permission from Sans et al.401 Copyright 2014, Royal Society of Chemistry.
Figure 14 Demonstration of an NMR system in the self-optimization of the condensation between aniline and p-Fluorobenzaldehyde. (A) Scheme of the utilized flow system is demonstrated showing the input streams of the reagents and trifluoroacetic acid (TFA) as catalyst, leading to the NMR spectrometer for analytical measurement. (B) Sample NMR spectrum from the optimization campaign, indicating characteristic product and reactant peaks. (C) Results of the optimization campaign to optimize objective function J over multiple iterations. Figure adapted with permission from Sans et al.401 Copyright 2014, Royal Society of Chemistry.CC BY-NC-ND · PubMed Central
Figure 15 Demonstration of an MS spectrometer in the self-optimization for the amidation of methyl nicotinate with aqueous MeNH2. The optimization considered the temperature, the reactant flow rate and MeNH2 equivalents to optimize the yield of the product drawn in brown. In the bottom row, a schematic overview over the used hardware is given (left) and the experimentally tested conditions in the optimization campaign (right). Figure adapted with permission from Holmes et al.403 Copyright 2016, Royal Society of Chemistry.
Figure 15 Demonstration of an MS spectrometer in the self-optimization for the amidation of methyl nicotinate with aqueous MeNH2. The optimization considered the temperature, the reactant flow rate and MeNH2 equivalents to optimize the yield of the product drawn in brown. In the bottom row, a schematic overview over the used hardware is given (left) and the experimentally tested conditions in the optimization campaign (right). Figure adapted with permission from Holmes et al.403 Copyright 2016, Royal Society of Chemistry.CC BY-NC-ND · PubMed Central
Figure 16 Modular reaction platform as proposed by Bédard et al. (A) The general four-step workflow consisting of the design of the synthesis path, loading the module, the self-optimization and obtaining the final results. (B) Construction of the modular system consisting of several bays that can contain multiple different modules and reagent inlets, leading to an in-line analytics system. (C) CAD representation of the LED reactor. (D) Schematic picture of the modular reactor platform. Figure adapted with permission from Bédard et al.264 Copyright 2018, American Association for the Advancement of Science.
Figure 16 Modular reaction platform as proposed by Bédard et al. (A) The general four-step workflow consisting of the design of the synthesis path, loading the module, the self-optimization and obtaining the final results. (B) Construction of the modular system consisting of several bays that can contain multiple different modules and reagent inlets, leading to an in-line analytics system. (C) CAD representation of the LED reactor. (D) Schematic picture of the modular reactor platform. Figure adapted with permission from Bédard et al.264 Copyright 2018, American Association for the Advancement of Science.CC BY-NC-ND · PubMed Central
Figure 17 Mobile robotic chemist as demonstrated by Burger et al. Schematic (left) shows the ten chemicals for which the concentrations were varied to optimize H2 production. Picture (right) portrays the (a) mobile robotic chemist with several workstations as well as (b) the general layout of the laboratory, with the mobile robot transferring the samples between the stations. Figure adapted with permission from Burger et al.11 Copyright 2020, Springer Nature.
Figure 17 Mobile robotic chemist as demonstrated by Burger et al. Schematic (left) shows the ten chemicals for which the concentrations were varied to optimize H2 production. Picture (right) portrays the (a) mobile robotic chemist with several workstations as well as (b) the general layout of the laboratory, with the mobile robot transferring the samples between the stations. Figure adapted with permission from Burger et al.11 Copyright 2020, Springer Nature.CC BY-NC-ND · PubMed Central
Figure 18 Demonstration of an example of a micro-droplet reactor for a Suzuki-Miyaura cross-coupling. (a) In the beginning of the reactor, droplets are generated with reagents selected by the optimization algorithm. Further down in the reactor, a base is added and the reaction mixture reacts further down in the reactor within the droplet, creating “batch in flow” conditions. After the reaction, the droplet is quenched and directed to an online analysis station. (b) Diagram shows the integration of the micro-droplet reactor into a self-optimization platform. Figure adapted with permission from reference Reizman et al.405 Copyright 2016, Royal Society of Chemistry.
Figure 18 Demonstration of an example of a micro-droplet reactor for a Suzuki-Miyaura cross-coupling. (a) In the beginning of the reactor, droplets are generated with reagents selected by the optimization algorithm. Further down in the reactor, a base is added and the reaction mixture reacts further down in the reactor within the droplet, creating “batch in flow” conditions. After the reaction, the droplet is quenched and directed to an online analysis station. (b) Diagram shows the integration of the micro-droplet reactor into a self-optimization platform. Figure adapted with permission from reference Reizman et al.405 Copyright 2016, Royal Society of Chemistry.CC BY-NC-ND · PubMed Central
Figure 19 Exemplary results for a Pareto front optimization of the E-factor and the space-time yield for an SNAr reaction. Identified Pareto optimal points are marked in orange, the interpolated Pareto front is drawn in red. In the left part of the Pareto front, a strong increase in space-time yield can be achieved by a minimal increase in E-factor, whereas the right side of the plot shows a strong increase in E-factor with only a small increase in space-time yield. Figure adapted with permission from reference Schweidtmann et al.414 Copyright 2018, Elsevier.
Figure 19 Exemplary results for a Pareto front optimization of the E-factor and the space-time yield for an SNAr reaction. Identified Pareto optimal points are marked in orange, the interpolated Pareto front is drawn in red. In the left part of the Pareto front, a strong increase in space-time yield can be achieved by a minimal increase in E-factor, whereas the right side of the plot shows a strong increase in E-factor with only a small increase in space-time yield. Figure adapted with permission from reference Schweidtmann et al.414 Copyright 2018, Elsevier.CC BY-NC-ND · PubMed Central
Figure 20 Reactions considered for self-optimization using deep RL. The bottom shows the different optimization runs for each of the reactions, respectively; comparing the algorithmic performance of RL (orange), one variable at a time (green) and the CMA-ES (blue). In all cases, the RL methods performed the optimization most efficiently. Figure adapted with permission from reference Zhou et al.334 Copyright 2017, American Chemical Society.
Figure 20 Reactions considered for self-optimization using deep RL. The bottom shows the different optimization runs for each of the reactions, respectively; comparing the algorithmic performance of RL (orange), one variable at a time (green) and the CMA-ES (blue). In all cases, the RL methods performed the optimization most efficiently. Figure adapted with permission from reference Zhou et al.334 Copyright 2017, American Chemical Society.CC BY-NC-ND · PubMed Central
Figure 21 Multi-Step Synthesis of the natural product Carpanone, where each reaction was considered as a separate self-optimization experiment. The synthesis was performed via a four-step route, consisting of an allylation of a phenol, a [3,3]-Claisen rearrangement, an isomerization and ultimately an oxidative dimerization. (A) shows the schematic overview of the reactor system for the first step with an HPLC/UV unit as an analytical unit, as well as the results of the optimization campaign (B) per experiment, and (C) as a function of the varied parameters. (D) Schematic experimental overview over the second step is provided, with an in-line NMR as an analytical unit. No further purification was necessary, since the product was obtained in 100% NMR yield. (E-F) The results of the optimization of the second reaction are shown. Figure adapted with permission from reference Cortés-Borda et al.413 Copyright 2018, American Chemical Society.
Figure 21 Multi-Step Synthesis of the natural product Carpanone, where each reaction was considered as a separate self-optimization experiment. The synthesis was performed via a four-step route, consisting of an allylation of a phenol, a [3,3]-Claisen rearrangement, an isomerization and ultimately an oxidative dimerization. (A) shows the schematic overview of the reactor system for the first step with an HPLC/UV unit as an analytical unit, as well as the results of the optimization campaign (B) per experiment, and (C) as a function of the varied parameters. (D) Schematic experimental overview over the second step is provided, with an in-line NMR as an analytical unit. No further purification was necessary, since the product was obtained in 100% NMR yield. (E-F) The results of the optimization of the second reaction are shown. Figure adapted with permission from reference Cortés-Borda et al.413 Copyright 2018, American Chemical Society.CC BY-NC-ND · PubMed Central
Figure 22 Examples of multi-step reactions by combining multiple single step self-optimization platforms. Top: Exemplary synthesis of PyOx ligands by Felpin and co-workers.415 Middle: Photo-catalytic two-step synthesis of a 2-oxazolidinone derivative performed by Jensen and co-workers.426 Bottom: Two-step synthesis of lidocaine by Ley and co-workers.116
Figure 22 Examples of multi-step reactions by combining multiple single step self-optimization platforms. Top: Exemplary synthesis of PyOx ligands by Felpin and co-workers.415 Middle: Photo-catalytic two-step synthesis of a 2-oxazolidinone derivative performed by Jensen and co-workers.426 Bottom: Two-step synthesis of lidocaine by Ley and co-workers.116CC BY-NC-ND · PubMed Central
Figure 23 Two-step synthesis of bupropion performed by Ley and co-workers. Initially, the two steps are optimized separately. The two optimized reactions are then combined with a work-up consisting of the addition of an aqueous sodium bisulfite solution and subsequent phase separation to yield one optimized reaction platform. Figure adapted with permission from reference Fitzpatrick et al.116 Copyright 2018, John Wiley and Sons.
Figure 23 Two-step synthesis of bupropion performed by Ley and co-workers. Initially, the two steps are optimized separately. The two optimized reactions are then combined with a work-up consisting of the addition of an aqueous sodium bisulfite solution and subsequent phase separation to yield one optimized reaction platform. Figure adapted with permission from reference Fitzpatrick et al.116 Copyright 2018, John Wiley and Sons.CC BY-NC-ND · PubMed Central
Figure 24 Demonstration of the telescoped synthesis of a Heck-coupling followed by a selective O-deprotection, as reported by Bourne and co-workers. (A) A schematic overview of the used reaction platform. In the initial reactor, the Heck coupling is performed. TsOH is added to the output of the first reactor to perform the selective O-deprotection. The output of both reactors is analyzed via one HPLC device connected via multipoint sampling (B). (C) The reaction pathway of the telescoped reaction. Analysis of the optimization data shows a different dominant pathway (3 → 5) compared to the pre hoc assumed one (3 → 4 → 5), underlining the utility of the optimization of a telescoped reaction. (D) Selected demonstrations of the HPLC chromatograms of the output of the first (top) and second (bottom) reactor. Figure adapted with permission from reference Clayton et al.430 Copyright 2022, John Wiley and Sons.
Figure 24 Demonstration of the telescoped synthesis of a Heck-coupling followed by a selective O-deprotection, as reported by Bourne and co-workers. (A) A schematic overview of the used reaction platform. In the initial reactor, the Heck coupling is performed. TsOH is added to the output of the first reactor to perform the selective O-deprotection. The output of both reactors is analyzed via one HPLC device connected via multipoint sampling (B). (C) The reaction pathway of the telescoped reaction. Analysis of the optimization data shows a different dominant pathway (3 → 5) compared to the pre hoc assumed one (3 → 4 → 5), underlining the utility of the optimization of a telescoped reaction. (D) Selected demonstrations of the HPLC chromatograms of the output of the first (top) and second (bottom) reactor. Figure adapted with permission from reference Clayton et al.430 Copyright 2022, John Wiley and Sons.CC BY-NC-ND · PubMed Central
Figure 25 Multi-step self-optimization of the synthesis of Sonidegib as proposed by Jensen and co-workers. The synthesis was performed over three steps: an SNAr reaction with a morpholine derivative as a nucleophile followed by a hydrogenation and an amide coupling. The latter two steps were performed as one telescoped synthesis. From a computer proposed and human-refined synthetic route, an approximate recipe was generated and continuous and categorical parameters were optimized in an automated robotic platform using a multi-objective BO algorithm (middle panel). Figure adapted with permission from reference Nambiar et al.427 Copyright 2022, American Chemical Society.
Figure 25 Multi-step self-optimization of the synthesis of Sonidegib as proposed by Jensen and co-workers. The synthesis was performed over three steps: an SNAr reaction with a morpholine derivative as a nucleophile followed by a hydrogenation and an amide coupling. The latter two steps were performed as one telescoped synthesis. From a computer proposed and human-refined synthetic route, an approximate recipe was generated and continuous and categorical parameters were optimized in an automated robotic platform using a multi-objective BO algorithm (middle panel). Figure adapted with permission from reference Nambiar et al.427 Copyright 2022, American Chemical Society.CC BY-NC-ND · PubMed Central
Figure 26 Experimental setup of the microfluidic platform for methane oxidation catalyst discovery. Left: Photograph of the 96-well plate for pre-mixing the catalyst components. Right: Schematic depiction of the microdroplet reactor containing a reaction droplet (left), and an indicator droplet (right). Figure adapted with permission from Kreutz et al.462 Copyright 2010, American Chemical Society.
Figure 26 Experimental setup of the microfluidic platform for methane oxidation catalyst discovery. Left: Photograph of the 96-well plate for pre-mixing the catalyst components. Right: Schematic depiction of the microdroplet reactor containing a reaction droplet (left), and an indicator droplet (right). Figure adapted with permission from Kreutz et al.462 Copyright 2010, American Chemical Society.CC BY-NC-ND · PubMed Central
Figure 27 Schematic workflow of the reaction discovery SDL developed by Cronin and co-workers. Figure reproduced with permission from Granda et al.471 Copyright 2020, Springer Nature.
Figure 27 Schematic workflow of the reaction discovery SDL developed by Cronin and co-workers. Figure reproduced with permission from Granda et al.471 Copyright 2020, Springer Nature.CC BY-NC-ND · PubMed Central
Figure 28 HTE platform used by Lin and co-workers, and the results of the closed-loop optimization. (A) Schematic and (B) picture of gantry-style SDL with multiple heads (C) to perform laser fabrication of microreactors, injection of precursors, and Joule heating synthesis of ZIFs. (D) Using a BO algorithm seeded with initial random samples of experimental parameters, the crystallinity is optimized with each iteration. (E) When compared to random sampling, BO achieves a higher crystallinity, measured by XRD. Figure adapted with permission from reference Xie et al.508 Copyright 2021, American Chemical Society.
Figure 28 HTE platform used by Lin and co-workers, and the results of the closed-loop optimization. (A) Schematic and (B) picture of gantry-style SDL with multiple heads (C) to perform laser fabrication of microreactors, injection of precursors, and Joule heating synthesis of ZIFs. (D) Using a BO algorithm seeded with initial random samples of experimental parameters, the crystallinity is optimized with each iteration. (E) When compared to random sampling, BO achieves a higher crystallinity, measured by XRD. Figure adapted with permission from reference Xie et al.508 Copyright 2021, American Chemical Society.CC BY-NC-ND · PubMed Central
Figure 29 Explored space of possible polyoxometalate crystals explored with each iteration. Duros et al. performed human-guided and random searches of the crystallization space as comparison for the algorithmic approach. The space is defined by the experimental parameters. Figure reproduced with permission from Duros et al.511 Copyright 2017, John Wiley and Sons.
Figure 29 Explored space of possible polyoxometalate crystals explored with each iteration. Duros et al. performed human-guided and random searches of the crystallization space as comparison for the algorithmic approach. The space is defined by the experimental parameters. Figure reproduced with permission from Duros et al.511 Copyright 2017, John Wiley and Sons.CC BY-NC-ND · PubMed Central
Figure 30 (a) Photo of the external of Adam, with Eve on the far right. (b) Photo of Adam’s robotic platform inside the system. Figures adapted with permission from Sparkes et al.546 Copyright 2010, Springer Nature.
Figure 30 (a) Photo of the external of Adam, with Eve on the far right. (b) Photo of Adam’s robotic platform inside the system. Figures adapted with permission from Sparkes et al.546 Copyright 2010, Springer Nature.CC BY-NC-ND · PubMed Central
Figure 31 Diagram of the early stage drug discovery pipeline. Robot scientist Eve37 is designed to automate the library screening, hit confirmation, and QSAR steps of the pipeline.
Figure 31 Diagram of the early stage drug discovery pipeline. Robot scientist Eve37 is designed to automate the library screening, hit confirmation, and QSAR steps of the pipeline.CC BY-NC-ND · PubMed Central
Figure 32 Automated pipeline for liver X receptor (LXR) agonist discovery. (A) The DL generative model was used to design candidate molecules. (B) A virtual reaction filter screened for synthesizability of the candidates. (C) Finally, select candidates were synthesized using a microfluidic platform. Figure reproduced with permission from Grisoni et al.550 Copyright 2021, American Association for the Advancement of Science.
Figure 32 Automated pipeline for liver X receptor (LXR) agonist discovery. (A) The DL generative model was used to design candidate molecules. (B) A virtual reaction filter screened for synthesizability of the candidates. (C) Finally, select candidates were synthesized using a microfluidic platform. Figure reproduced with permission from Grisoni et al.550 Copyright 2021, American Association for the Advancement of Science.CC BY-NC-ND · PubMed Central
Figure 33 (left) Schematic of the integrated design, synthesis, and screening platform illustrating the fully automated processes implemented for closed-loop drug discovery. Following initiation of the process the system completes multiple iterations of design, synthesis, and screening without manual intervention. (right) Schematic showing the continuous fluidic path taken by reagents and products on the platform. Figure reprinted with permission from Desai et al.555 Copyright 2013, American Chemical Society.
Figure 33 (left) Schematic of the integrated design, synthesis, and screening platform illustrating the fully automated processes implemented for closed-loop drug discovery. Following initiation of the process the system completes multiple iterations of design, synthesis, and screening without manual intervention. (right) Schematic showing the continuous fluidic path taken by reagents and products on the platform. Figure reprinted with permission from Desai et al.555 Copyright 2013, American Chemical Society.CC BY-NC-ND · PubMed Central
Figure 34 (left) Scheme of the adopted closed-loop optimization workflow. Material flow (continuous lines) and information flow (dashed lines) are reported. Ingredients are mixed following the suggested recipes in robot R1, processed, and analyzed with a combination of in-line automated operations and offline manual analyses. The results are then collected and processed by the algorithm to suggest a new set of experiments to run for the next iteration. (right) Image of the experimental setup based on the two formulation robots, R1 and R2. The picture shows the actual experimental setup as used for the experiments. Automated syringe pumps (B) are connected to feeding bottles (A) to dispense ingredients to different vials located on the rotating wheel (C) of robot R1. Samples are then moved to the offline incubator for processing and placed in robot R2, where image collection (D), turbidity (F), and pH (G) analyses can be run. The platforms are controlled by the PC (E), where data is stored and fed to the algorithm for the generation of the next iteration. Figure adapted with permission from Cao et al.560 Copyright 2021, Elsevier.
Figure 34 (left) Scheme of the adopted closed-loop optimization workflow. Material flow (continuous lines) and information flow (dashed lines) are reported. Ingredients are mixed following the suggested recipes in robot R1, processed, and analyzed with a combination of in-line automated operations and offline manual analyses. The results are then collected and processed by the algorithm to suggest a new set of experiments to run for the next iteration. (right) Image of the experimental setup based on the two formulation robots, R1 and R2. The picture shows the actual experimental setup as used for the experiments. Automated syringe pumps (B) are connected to feeding bottles (A) to dispense ingredients to different vials located on the rotating wheel (C) of robot R1. Samples are then moved to the offline incubator for processing and placed in robot R2, where image collection (D), turbidity (F), and pH (G) analyses can be run. The platforms are controlled by the PC (E), where data is stored and fed to the algorithm for the generation of the next iteration. Figure adapted with permission from Cao et al.560 Copyright 2021, Elsevier.CC BY-NC-ND · PubMed Central
Figure 35 Overview of the BioMAP platform for biologic formulation. Biologic formulation is performed entirely through autonomous workflows. Multisource data from physical and biological experiments are exploited by deep neural networks to map complex structure-function landscapes and inform downstream design campaigns. Figure reproduced with permission from Tamasi et al.566 Copyright 2022, Elsevier.
Figure 35 Overview of the BioMAP platform for biologic formulation. Biologic formulation is performed entirely through autonomous workflows. Multisource data from physical and biological experiments are exploited by deep neural networks to map complex structure-function landscapes and inform downstream design campaigns. Figure reproduced with permission from Tamasi et al.566 Copyright 2022, Elsevier.CC BY-NC-ND · PubMed Central
Figure 36 Schematic of the SAMPLE workflow. An intelligent agent (left) learns sequence-function relationships and designs proteins to test hypotheses. The agent sends designed proteins to a laboratory environment (right) that performs fully automated gene assembly, protein expression, and biochemical characterization, and sends the resulting data back to the agent, which refines its understanding of the system and repeats the process. Figure reproduced with permission from Rapp et al.579 Copyright 2024, Springer Nature.
Figure 36 Schematic of the SAMPLE workflow. An intelligent agent (left) learns sequence-function relationships and designs proteins to test hypotheses. The agent sends designed proteins to a laboratory environment (right) that performs fully automated gene assembly, protein expression, and biochemical characterization, and sends the resulting data back to the agent, which refines its understanding of the system and repeats the process. Figure reproduced with permission from Rapp et al.579 Copyright 2024, Springer Nature.CC BY-NC-ND · PubMed Central
Figure 37 A spider plot illustrating twelve performance properties for structural applications. Two example alloys are evaluated along the twelve axes. Figure reproduced with permission from Miracle et al.594 Copyright 2017, Elsevier.
Figure 37 A spider plot illustrating twelve performance properties for structural applications. Two example alloys are evaluated along the twelve axes. Figure reproduced with permission from Miracle et al.594 Copyright 2017, Elsevier.CC BY-NC-ND · PubMed Central
Figure 38 HT-READ incorporates physics- and thermodynamics-based models with automated synthesis and high-throughput characterization methods for virtual alloy screening. Figure reproduced with permission from Vecchio et al.595 Copyright 2021, Elsevier.
Figure 38 HT-READ incorporates physics- and thermodynamics-based models with automated synthesis and high-throughput characterization methods for virtual alloy screening. Figure reproduced with permission from Vecchio et al.595 Copyright 2021, Elsevier.CC BY-NC-ND · PubMed Central
Figure 39 A robotic platform with custom tooling for processing and testing adhesive specimens where batches of next experiments are guided via BO. (a) The automated platform with N9 arm capable of picking up and using various attachments. (b) The testing head used to characterize the force required to break the adhesive material. (c) The workflow of the SDL, note that adhesives were created and loaded manually. Figure reproduced with permission from Rooney et al.306 Copyright 2022, Royal Society of Chemistry.
Figure 39 A robotic platform with custom tooling for processing and testing adhesive specimens where batches of next experiments are guided via BO. (a) The automated platform with N9 arm capable of picking up and using various attachments. (b) The testing head used to characterize the force required to break the adhesive material. (c) The workflow of the SDL, note that adhesives were created and loaded manually. Figure reproduced with permission from Rooney et al.306 Copyright 2022, Royal Society of Chemistry.CC BY-NC-ND · PubMed Central
Figure 40 Schematic of SDL from Sargent and co-workers, studying the crystallization of perovskites. (A) A diagram of experimental setup used for antisolvent vapor-assisted crystallization. (B) A tray with 96 containers, each containing 3 drops of precursor solution, and a well of antisolvent, which were then sealed for the crystallization. (C) Microscopy images over the course of 3 hours of the crystallization. (D) The SDL workflow. (E) A schematic ML method used. The convolution layers perform operations on spatially connected pixels of the images, finally being flattened and put through a fully connected layer to output a classification. Figure adapted with permission from Kirman et al.617 Copyright 2020, Elsevier.
Figure 40 Schematic of SDL from Sargent and co-workers, studying the crystallization of perovskites. (A) A diagram of experimental setup used for antisolvent vapor-assisted crystallization. (B) A tray with 96 containers, each containing 3 drops of precursor solution, and a well of antisolvent, which were then sealed for the crystallization. (C) Microscopy images over the course of 3 hours of the crystallization. (D) The SDL workflow. (E) A schematic ML method used. The convolution layers perform operations on spatially connected pixels of the images, finally being flattened and put through a fully connected layer to output a classification. Figure adapted with permission from Kirman et al.617 Copyright 2020, Elsevier.CC BY-NC-ND · PubMed Central
Figure 41 Schematic of the microfluidics platform, and the results of the SDL optimization from deMello and co-workers. (top) A schematic of the microfluidic SDL, with the channels etched onto a glass substrate. Precursors flowed through the chip using syringes, and the resulting NPs were characterized by emission spectra. (bottom) The dissatisfaction coefficient was calculated from the characterization of the NPs, and was minimized using SNOBFIT. The minimum in the dissatisfaction coefficient is shown as a function of experimental parameters (flow rate of precursors). Figure adapted with permission from Krishnadasan et al.34 Copyright 2001, Royal Society of Chemistry.
Figure 41 Schematic of the microfluidics platform, and the results of the SDL optimization from deMello and co-workers. (top) A schematic of the microfluidic SDL, with the channels etched onto a glass substrate. Precursors flowed through the chip using syringes, and the resulting NPs were characterized by emission spectra. (bottom) The dissatisfaction coefficient was calculated from the characterization of the NPs, and was minimized using SNOBFIT. The minimum in the dissatisfaction coefficient is shown as a function of experimental parameters (flow rate of precursors). Figure adapted with permission from Krishnadasan et al.34 Copyright 2001, Royal Society of Chemistry.CC BY-NC-ND · PubMed Central
Figure 42 Schematic of the artificial chemist developed by Abolhasani and co-workers. Variations of this platform were used in future work from the group. (A) Continuous flow systems powered by a series of syringe pumps. A custom in situ spectrometer system was used to monitor the absorption and emission spectra. (B) A schematic of the workflow of the artificial chemist. Starting from random experiments, the flow synthesis and characterization were followed by a boosted ensemble of neural networks. Figure reproduced with permission from Epps et al.627 Copyright 2020, John Wiley and Sons.
Figure 42 Schematic of the artificial chemist developed by Abolhasani and co-workers. Variations of this platform were used in future work from the group. (A) Continuous flow systems powered by a series of syringe pumps. A custom in situ spectrometer system was used to monitor the absorption and emission spectra. (B) A schematic of the workflow of the artificial chemist. Starting from random experiments, the flow synthesis and characterization were followed by a boosted ensemble of neural networks. Figure reproduced with permission from Epps et al.627 Copyright 2020, John Wiley and Sons.CC BY-NC-ND · PubMed Central
Figure 43 MAOSIC platform with automated synthesis and characterization connected via a cloud platform. Synthesis was performed with a microfluidic reactor, followed by an in situ spectrometer. The samples were then transferred by a robotic arm to the CD spectrometer. All operation, data management, and optimization algorithms were done remotely and collaboratively in the cloud. Figure reproduced with permission from Li et al.119 Copyright 2020, Springer Nature.
Figure 43 MAOSIC platform with automated synthesis and characterization connected via a cloud platform. Synthesis was performed with a microfluidic reactor, followed by an in situ spectrometer. The samples were then transferred by a robotic arm to the CD spectrometer. All operation, data management, and optimization algorithms were done remotely and collaboratively in the cloud. Figure reproduced with permission from Li et al.119 Copyright 2020, Springer Nature.CC BY-NC-ND · PubMed Central
Figure 44 Schematic describing the closed-loop optimization campaign for discovery of organic solid-state lasers. (A) Multiple laboratories from across the globe ran asynchronous experiments, with experimental planning and results coordinated through an online server. (B) The parallel asynchronous optimization visualized as a timeline, demonstrating how 4 different labs can coordinate experiments. (C) The organic molecules were created and the lasing performances were approximated using the emission gain factor. After closed-loop automated discovery, the best molecules were used to test thin-film device performance. (D) The chemical space was defined through the use of cap, bridge, and core fragments, linked together via Suzuki-Miyaura coupling reactions. Figure reproduced with permission from Strieth-Kalthoff et al.638 Copyright 2024, American Association for the Advancement of Science.
Figure 44 Schematic describing the closed-loop optimization campaign for discovery of organic solid-state lasers. (A) Multiple laboratories from across the globe ran asynchronous experiments, with experimental planning and results coordinated through an online server. (B) The parallel asynchronous optimization visualized as a timeline, demonstrating how 4 different labs can coordinate experiments. (C) The organic molecules were created and the lasing performances were approximated using the emission gain factor. After closed-loop automated discovery, the best molecules were used to test thin-film device performance. (D) The chemical space was defined through the use of cap, bridge, and core fragments, linked together via Suzuki-Miyaura coupling reactions. Figure reproduced with permission from Strieth-Kalthoff et al.638 Copyright 2024, American Association for the Advancement of Science.CC BY-NC-ND · PubMed Central
Figure 45 Configuration of Ada robot used by Berlinguette and co-workers to study the conductivity and annealing temperature of Pd films. A similar platform was used for the optimization of thin film fabrication. The modular nature of Ada allows for different instruments to be connected to the platform. (a) A robotic arm transferred the sample, after thin film deposition, from the 4-axis robot platform to the XRF microscope, and back to the 4-axis robot for characterization. (b) An expanded diagram of the 4-axis robot platform, as well as the various stages of operation of the automated synthesis and characterization. Figure reproduced with permission from MacLeod et al.650 Copyright 2022, Springer Nature.
Figure 45 Configuration of Ada robot used by Berlinguette and co-workers to study the conductivity and annealing temperature of Pd films. A similar platform was used for the optimization of thin film fabrication. The modular nature of Ada allows for different instruments to be connected to the platform. (a) A robotic arm transferred the sample, after thin film deposition, from the 4-axis robot platform to the XRF microscope, and back to the 4-axis robot for characterization. (b) An expanded diagram of the 4-axis robot platform, as well as the various stages of operation of the automated synthesis and characterization. Figure reproduced with permission from MacLeod et al.650 Copyright 2022, Springer Nature.CC BY-NC-ND · PubMed Central
Figure 46 Compounds used in quaternary OPV films systems. Two mixes were studied by Langner et al.:651 (1) P3HT, PBQ-QF, PCBM, and oIDTBr, and (2) P3HT, PTB7-Th, PCBM, and oIDTBr.
Figure 46 Compounds used in quaternary OPV films systems. Two mixes were studied by Langner et al.:651 (1) P3HT, PBQ-QF, PCBM, and oIDTBr, and (2) P3HT, PTB7-Th, PCBM, and oIDTBr.CC BY-NC-ND · PubMed Central
Figure 47 Summary of AMANDA Line One from the SDL work of Brabec and co-workers. (A) the OPV materials of interest, with PM6 as the donor, and Y6 as the acceptor. (B) The device structure, with the active layer containing the D:A compounds. (C) Some of the dimensions varied in the device fabrication. (D) Picture of the automated layer deposition platform used to fabricate the devices. (E) Schematic of the automated characterization methods, with additional off-line studies of the degradation. (F) Plots of the current density and absorption spectra measurements of multiple solar cells, demonstrating the reproducibility. (E) The workflow of AMANDA Line One. Figure reproduced with permission from Du et al.653 Copyright 2021, Elsevier.
Figure 47 Summary of AMANDA Line One from the SDL work of Brabec and co-workers. (A) the OPV materials of interest, with PM6 as the donor, and Y6 as the acceptor. (B) The device structure, with the active layer containing the D:A compounds. (C) Some of the dimensions varied in the device fabrication. (D) Picture of the automated layer deposition platform used to fabricate the devices. (E) Schematic of the automated characterization methods, with additional off-line studies of the degradation. (F) Plots of the current density and absorption spectra measurements of multiple solar cells, demonstrating the reproducibility. (E) The workflow of AMANDA Line One. Figure reproduced with permission from Du et al.653 Copyright 2021, Elsevier.CC BY-NC-ND · PubMed Central
Figure 48 Electrolab is an automated electrochemistry platform that can dispense solutions onto electrochemical cells and run cyclic voltammograms, all without human intervention. (A) A schematic and picture of the Electrolab gantry-style SDL. (B) The control system of the Electrolab. (C) The vital electrochemical modules, with a microfabricated “eChip” electrode array for CV scans. (D) Fluidic nozzle system controlled by the gantry that can dispense and dispose fluids, rinse and flush with solvent, and dry and sparge with Ar gas. Figure reproduced with permission from Oh et al.661 Copyright 2023, Elsevier.
Figure 48 Electrolab is an automated electrochemistry platform that can dispense solutions onto electrochemical cells and run cyclic voltammograms, all without human intervention. (A) A schematic and picture of the Electrolab gantry-style SDL. (B) The control system of the Electrolab. (C) The vital electrochemical modules, with a microfabricated “eChip” electrode array for CV scans. (D) Fluidic nozzle system controlled by the gantry that can dispense and dispose fluids, rinse and flush with solvent, and dry and sparge with Ar gas. Figure reproduced with permission from Oh et al.661 Copyright 2023, Elsevier.CC BY-NC-ND · PubMed Central
Figure 49 Illustration of CASH (Connected, Autonomous, Shared, and High-throughput) (A) proposed by Shimizu et al. (B) The system synthesizes thin films using deposition conditions commanded by BO, after which the film's resistance is evaluated. Adapted with permission from Shimizu et al.677 Copyright 2020, American Institute of Physics Publishing.
Figure 49 Illustration of CASH (Connected, Autonomous, Shared, and High-throughput) (A) proposed by Shimizu et al. (B) The system synthesizes thin films using deposition conditions commanded by BO, after which the film's resistance is evaluated. Adapted with permission from Shimizu et al.677 Copyright 2020, American Institute of Physics Publishing.CC BY-NC-ND · PubMed Central
Figure 50 (left) Diagram illustrating the steps of the data-driven high-throughput automated robotic experiments designed to assess the Coulombic efficiencies of multi-component electrolyte additives.(right) Graph illustrating the average Coulombic efficiency (CE) achieved by the leading 5 samples over the course of the experiments. The x-axis represents the experiment iteration, distinguishing between random search (black), hill climbing with the top 10 samples (red), hill climbing with the top 50 samples (green), and BO (blue). Figure adapted with permission from Matsuda et al.686 Copyright 2022, Elsevier.
Figure 50 (left) Diagram illustrating the steps of the data-driven high-throughput automated robotic experiments designed to assess the Coulombic efficiencies of multi-component electrolyte additives.(right) Graph illustrating the average Coulombic efficiency (CE) achieved by the leading 5 samples over the course of the experiments. The x-axis represents the experiment iteration, distinguishing between random search (black), hill climbing with the top 10 samples (red), hill climbing with the top 50 samples (green), and BO (blue). Figure adapted with permission from Matsuda et al.686 Copyright 2022, Elsevier.CC BY-NC-ND · PubMed Central
Figure 51 Schematic illustration of the high-throughput synthesis protocol for DESs used by Pozzo and co-workers. (a) The chemical space is defined by the QASs and HBDs used. (b) Solutions were prepared for the liquid handling robot (c). (d) The DES samples in 48-well plates were transferred to a dehydrator (e). (f) The samples were heated, and then (g) placed in a vacuum. (h) Final parallel analysis of the formed DESs. Reproduced with permission from Rodriguez et al.690 Copyright 2016, Institution of Chemical Engineers (IChemE) and the Royal Society of Chemistry.
Figure 51 Schematic illustration of the high-throughput synthesis protocol for DESs used by Pozzo and co-workers. (a) The chemical space is defined by the QASs and HBDs used. (b) Solutions were prepared for the liquid handling robot (c). (d) The DES samples in 48-well plates were transferred to a dehydrator (e). (f) The samples were heated, and then (g) placed in a vacuum. (h) Final parallel analysis of the formed DESs. Reproduced with permission from Rodriguez et al.690 Copyright 2016, Institution of Chemical Engineers (IChemE) and the Royal Society of Chemistry.CC BY-NC-ND · PubMed Central
Figure 52 Diagram illustrating the SDL setup designed for the electrochemical deposition and evaluation of the performance of OER catalysts made of amorphous mixed-metal oxide. (a) The EMAP (Electrocatalysis material acceleration platform) is comprised of (1) a robotic arm with an integrated pipette holder, (2) weigh scale and gripper, (3) liquid dispensing carousel, (4) slide hotel, (5) syringe pumps, (6) pipette tip holder, (7) vial rack, (8) slide gripper and (9) automated electrochemical cell. (b) Zoomed-in view of automated electrochemical cells. Adapted with permission from Fatehi et al.707
Figure 52 Diagram illustrating the SDL setup designed for the electrochemical deposition and evaluation of the performance of OER catalysts made of amorphous mixed-metal oxide. (a) The EMAP (Electrocatalysis material acceleration platform) is comprised of (1) a robotic arm with an integrated pipette holder, (2) weigh scale and gripper, (3) liquid dispensing carousel, (4) slide hotel, (5) syringe pumps, (6) pipette tip holder, (7) vial rack, (8) slide gripper and (9) automated electrochemical cell. (b) Zoomed-in view of automated electrochemical cells. Adapted with permission from Fatehi et al.707CC BY-NC-ND · PubMed Central
Figure 53 A visual representation of the automated battery assembly system (AutoBASS) is depicted, showcasing part trays designed for the assembly of CR2023 cells. The components are selected from these trays and positioned onto the assembly post using a gentle silicone suction cup attached to robot A. The cells are constructed with a downward-facing anode cap, and gripper B is responsible for flipping them. Gripper B also moves the filled and assembled cells to the crimper. The extraction from the crimper is facilitated by a magnetic pickup mechanism. Figure reproduced with permission from Zhang et al.709 Copyright 2022, Royal Society of Chemistry.
Figure 53 A visual representation of the automated battery assembly system (AutoBASS) is depicted, showcasing part trays designed for the assembly of CR2023 cells. The components are selected from these trays and positioned onto the assembly post using a gentle silicone suction cup attached to robot A. The cells are constructed with a downward-facing anode cap, and gripper B is responsible for flipping them. Gripper B also moves the filled and assembled cells to the crimper. The extraction from the crimper is facilitated by a magnetic pickup mechanism. Figure reproduced with permission from Zhang et al.709 Copyright 2022, Royal Society of Chemistry.CC BY-NC-ND · PubMed Central
Figure 54 A flow reactor setup for bicarbonate reduction converts captured CO2 into CO, which can be converted into other fuels. (A) A schematic demonstrating the electrochemical flow cell experiment, with corresponding reactions. (B) An expanded view of the flow cell within the stainless steel housing. (C) The flow plates used for the cathode and anode. Figure reproduced with permission from Li et al.712 Copyright 2019, Elsevier.
Figure 54 A flow reactor setup for bicarbonate reduction converts captured CO2 into CO, which can be converted into other fuels. (A) A schematic demonstrating the electrochemical flow cell experiment, with corresponding reactions. (B) An expanded view of the flow cell within the stainless steel housing. (C) The flow plates used for the cathode and anode. Figure reproduced with permission from Li et al.712 Copyright 2019, Elsevier.CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

1 more claim about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 732 references

Research works printed but not linked: 25

  1. 2011, Laboratory Safety Guidance
  2. John Wiley & Sons, Inc. SpectraBase . https://spectrabase.com/ (accessed 2023-10-10).
  3. Williams C. K., 2006, Gaussian Processes for Machine Learning, 2
  4. MacQueen J., 1967, Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, Volume 1: Statistics, 5, 281
  5. Mauri A., 2006, MATCH Commun. Math. Comput. Chem., 56, 237
  6. Bengio Y., 2023, J. Mach. Learn. Res., 24, 1
  7. Bergstra J., 2012, J. Mach. Learn. Res., 13, 281
  8. Fisher R. A., 1960, The Design of Experiments
  9. 10.1016/0098-1354(90)87072-W · DOI
  10. Kikkawa N., 2024, J. Mach. Learn. Res., 25, 1
  11. 10.48550/arXiv.2303.01560 · arXiv · doi:10.48550/arxiv.2303.01560
  12. Bakshy, E.; Dworkin, L.; Karrer, B.; Kashin, K.; Letham, B.; Murthy, A.; Singh, S. AE: A Domain-Agnostic Platform for Adaptive Experimentation. In Conference on neural information processing systems ; 2018; pp 1−8.
  13. Paszke A., 2019, Advances in Neural Information Processing Systems 32, 8024
  14. Gardner J., 2018, Adv. Neural Inf. Process. Syst., 31
  15. Lindauer M., 2022, J. Mach. Learn. Res., 23, 1
  16. Lindauer M., 2022, Journal of Machine Learning Research, 23, 1
  17. 10.1109/4235.585893 · DOI · Paper page
  18. 10.1145/1377612.1377613 · DOI · Paper page
  19. Rozycki C., 1993, Chem. Anal., 38, 681
  20. Labes R., 2020, Chem. Today., 38
  21. 10.1063/5.0048164 · DOI · Paper page
  22. 10.1145/1377612.1377613 · DOI · Paper page
  23. 10.1002/adma.202207070 · DOI
  24. 10.1021/acscentsci.8b00307 · DOI · Paper page
  25. 10.1126/science.aat2663 · DOI · Paper page

Cited by

See all 769

Metadata

Authors
Tom, G., Schmid, S. P., Baird, S. G., Cao, Y., Darvish, K., Hao, H., Lo, S., Pablo-García, S., Rajaonson, E. M., Skreta, M., Yoshikawa, N., Corapi, S., Akkoc, G. D., Strieth-Kalthoff, F., Seifrid, M., & Aspuru-Guzik, A.
Affiliations
  • University of Toronto (CA)
  • ETH Zurich (CH)
  • Acceleration Consortium, 80 St. George St, Toronto, Ontario M5S 3H6, Canada
  • Department of Chemistry, University of Toronto, 80 St. George St, Toronto, Ontario M5S 3H6, Canada
  • Forschungszentrum Jülich (DE)
  • North Carolina State University (US)
Corresponding author
  • Tom, Gary
  • Seifrid, Martin
Venue
Chemical Reviews
Published
2024-08-13
Volume
124
Issue
16
Pages
9633-9732
ISSN
  • 0009-2665 (ISSN-L)
  • 1520-6890
OpenAlex
W4401535000
Cited by
707
References
698
Citations by year
2026: 403 · 2025: 279 · 2024: 23
FWCI
16.77
Type
journal article
Language
en
Open access
Locations
  • Chemical Reviews (publishedVersion, cc-by-nc-nd)
  • ChemRxiv (acceptedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion)
  • Repository for Publications and Research Data (ETH Zurich) (submittedVersion, cc-by-nc-nd)
  • ETH Zürich Research Collection
Funders
  • Bundesministerium f?r Bildung und Forschung (03HY108A)
  • Defense Advanced Research Projects Agency (HR00111920027)
  • Canada Research Chairs
  • Office of Science (DE-SC0023454)
  • Schmidt Futures
  • Canada First Research Excellence Fund
  • University of Minnesota
  • Natural Sciences and Engineering Research Council of Canada
  • Canadian Institute for Advanced Research
  • Vector Institute
  • Government of Ontario
  • Mitacs
Topics
  • Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
  • Social Sciences › Education › Various Chemistry Research Topics
  • Physical Sciences › Chemical Engineering › Fluid Flow and Transfer Processes › Innovative Microfluidic and Catalytic Techniques Innovation
Keywords
self-driving laboratories, laboratory automation, materials discovery, chemistry, drug discovery, genomics, high-throughput experimentation, research automation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Self-Driving Laboratories for Chemistry and Materials Science. https://app.cochat.ai/papers/10.1021/acs.chemrev.4c00055
The paper
Tom, G., Schmid, S. P., Baird, S. G., Cao, Y., Darvish, K., Hao, H., Lo, S., Pablo-García, S., Rajaonson, E. M., Skreta, M., Yoshikawa, N., Corapi, S., Akkoc, G. D., Strieth-Kalthoff, F., Seifrid, M., & Aspuru-Guzik, A. (2024). Self-Driving Laboratories for Chemistry and Materials Science. Chemical Reviews. https://doi.org/10.1021/acs.chemrev.4c00055

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.