The automation of chemical synthesis processes through mobile robots is making significant progress, with the potential to accelerate and make the discovery of new chemical compounds more efficient. In this article, we examine a recent study conducted by Tianwei Dai, Sriram Vijayakrishnan, Filip T. Szczypiński, Jean-François Ayme, Ehsan Simaei, Thomas Fellowes, Rob Clowes, Lyubomir Kotopanov, Caitlin E. Shields, Zhengxue Zhou, John W. Ward, and Andrew I. Cooper, in collaboration with the University of Liverpool and other institutions. This work presents an autonomous laboratory that performs exploratory chemistry processes in a fully automated manner, simulating the decision-making process of human chemists. The innovative platform finds applications in the synthesis of structurally diverse compounds, supramolecular chemistry, and photocatalytic synthesis, opening new perspectives for chemical research.
Challenges of Exploratory Chemistry
Exploratory chemistry aims to discover new reactions and compounds with unique properties and applications. Unlike more traditional and systematic chemistry, exploratory chemistry often encounters situations where results are unpredictable, and optimal reaction conditions can vary significantly. The ability to adapt to these unexpected outcomes and make the best use of the diversity of chemical reactions represents one of the main challenges for those seeking to automate these processes.
Autonomous laboratories, with robots operating without constant supervision, represent an important opportunity to speed up the discovery of new chemical compounds. However, to achieve this level of autonomy, robots must not only be able to conduct experiments but also interpret analytical data and make decisions based on that data. This represents a significant challenge, as exploratory chemistry yields products with diverse and sometimes unpredictable characteristics.
In traditional exploratory chemistry, the chemist acts as the central decision-maker, continuously adapting experiments based on the data collected. The interpretation of information from various analytical techniques, such as mass spectrometry (MS) and nuclear magnetic resonance (NMR), is a key element of this process. These instruments allow detailed data to be obtained on molecular structure, purity characteristics, and possible reactions. Humans are capable of interpreting this information intuitively and making complex contextual decisions, such as deciding whether a product is valid, if a reaction needs to be repeated, or if the results are worth further exploration.
Robotic systems must therefore be programmed to manage large volumes of complex and multimodal data, such as those provided by MS and NMR analyses. The main difficulty lies in handling data variability and making decisions based on predetermined rules without falling into errors caused by overgeneralizations. For example, some molecules may produce very complex MS spectra but simple NMR spectra, or vice versa, complicating the automated product quality assessment phase.
Another challenging element is the open-ended nature of exploratory chemistry. Unlike targeted synthesis chemistry, where success can be measured in terms of yield or purity of the desired product, exploratory chemistry often aims to identify new molecules without necessarily knowing what product will be obtained in advance. In these conditions, the lack of a clear "success score" complicates automation. Autonomous synthesis, therefore, requires decision algorithms that are flexible enough to adapt to unstructured and open situations but also robust enough to avoid wasting resources on unproductive reactions.
To address these challenges, a modular approach has been developed that combines mobile robots and distributed analytical instruments, together with an algorithmic decision-maker based on heuristic rules. This approach allows data from different analytical sources, such as NMR and UPLC-MS, to be processed, making autonomous decisions on which reactions to pursue. The heuristics used to make these decisions are simply algorithmic representations of expert knowledge, similar to those a skilled chemist would use in the lab. These rules can be adjusted to respond to specific characteristics of the reactions, making the platform versatile and capable of tackling a wide range of problems.
The system described in the study was designed to be adaptable. This means that, with the addition of further analytical instruments and synthesis modules, the capabilities of the autonomous laboratory can be expanded, thereby improving the ability to characterize an increasingly wide range of products. The modularity also allows for the integration of new technologies without the need to completely redesign the system, ensuring that the platform can evolve with technological progress.
Robotics and Exploratory Chemistry
The autonomous platform has been designed in a modular fashion to transfer samples between different synthesis and analysis phases using mobile robots. The laboratory integrates an automated synthesis platform Chemspeed ISynth, a UPLC-MS (Ultra High-Performance Liquid Chromatography-Mass Spectrometry) spectrometer, and an 80 MHz benchtop NMR.
The Chemspeed ISynth system, the core of the synthesis platform, can handle up to 96 simultaneous reactions thanks to its parallel reactors, each equipped with temperature control (up to 200°C) and variable speed stirring capabilities. This provides great flexibility in experiment design and the ability to explore different reaction conditions in parallel. The ability to conduct simultaneous reactions dramatically reduces the time required to optimize a reaction, increasing laboratory productivity.
The UPLC-MS spectrometer, an essential component for analyzing reaction products, offers high resolution to separate compounds and identify products with high sensitivity. The system is equipped with a pump capable of operating at pressures up to 15,000 psi, ensuring efficient separation even for complex mixtures. Mass detection is performed using a quadrupole detector with a sensitivity that allows the detection of compounds at concentrations in the nanogram per milliliter range. This level of precision is crucial to ensure that even trace products can be correctly identified and characterized.
The benchtop NMR used is a Bruker Fourier80, with a frequency of 80 MHz for the hydrogen nucleus (1H). Despite being a compact system, the Fourier80 can provide fundamental structural data on synthesized molecules, allowing analysis of product conformation and integrity. The NMR is equipped with solvent suppression capabilities, which reduces interference from solvents used in reactions, improving the quality of the obtained spectrum.
The mobile robots used to transport samples between various instruments are KUKA robots equipped with precision encoders, laser scanners, and force sensors, allowing them to navigate autonomously in the laboratory with an accuracy of ±0.12 mm and an orientation precision of θ ± 0.005°. Each robot is configured with a multifunction gripper system capable of handling different types of containers, from reaction vials to analytical samples, ensuring safe and precise handling of materials. The average time to transfer a sample between two instruments is about 2 minutes, helping to minimize downtime between synthesis and analysis phases.
Another key aspect of the platform is the central control system, called the Intelligent Automation System Control Panel (IAS-CP), which coordinates all laboratory activities. The IAS-CP communicates with various modules and instruments through a ZeroMQ communication protocol, enabling distributed control and reducing bottlenecks in the automation process. Every action is tracked and recorded in a central database, allowing for complete traceability of operations and subsequent analysis to continuously improve platform performance.
Decisions regarding subsequent synthesis phases are made by a decision algorithm that evaluates data from the analytical instruments using a series of heuristic rules. This approach allows human decision-making to be emulated in an automated and scalable way, ensuring that the most promising reactions are carried forward without the need for human intervention.
Diverse Structural Chemistry and Automation
The platform makes a significant contribution to innovation in drug discovery. Designed to perform multiple reactions in parallel, it allows for rapid and efficient exploration of a wide range of experimental conditions. This approach not only accelerates the identification of potentially useful compounds but also optimizes the resources employed in pharmacological research, making the discovery process more sustainable and effective.
The platform carried out the parallel synthesis of three ureas and three thioureas through the condensation of amines with isocyanates and isothiocyanates, followed by analysis using UPLC-MS and NMR. Each reaction was conducted in an average volume of 1-5 mL, with a reagent dosing precision of ±0.01 mL, ensuring a high level of reproducibility. Reaction temperatures were controlled in a range from 25°C to 120°C, and stirring conditions varied from 500 to 1500 rpm, depending on viscosity.
The decision algorithm used data from 64 NMR scans for each sample, obtaining a detailed profile of the molecular structure of the products. UPLC-MS mass spectrometry enabled the identification of target products with a sensitivity of up to 10 ng/mL, ensuring that even compounds formed in small amounts could be detected. Once positive products were identified, the system automatically triggered the scale-up process, increasing the reaction volume from 1 mL to 10 mL for further studies.
Among the reactions performed by the platform, a Sonogashira cross-coupling reaction stands out, using a palladium catalyst under anaerobic conditions, achieving a 78% yield after 8 hours of reaction. The platform also conducted a copper-catalyzed azide-alkyne cycloaddition (CuAAC) reaction with an efficiency of 92%, demonstrating its ability to manage complex reactions requiring specific conditions, such as the presence of inert gases. These results highlight the reliability and versatility of the system in managing advanced chemical processes.
The autonomous system operated without human intervention for four consecutive days, processing over 50 different reactions and requiring only the replacement of reagents and consumables, an activity that could also be automated in the future. During these four days, the system analyzed over 500 samples and generated more than 5 GB of analytical data, which were processed in real-time by the central control system to make decisions on subsequent synthesis phases.
This ability to operate autonomously for extended periods, processing large amounts of data in real-time and making data-driven decisions, represents a significant step forward compared to traditional chemical synthesis methods, where human involvement is constant and decision-making is often limited by the speed at which results can be analyzed.
Autonomous Supramolecular Synthesis
The autonomous laboratory was also employed in supramolecular synthesis, a branch of chemistry focused on creating complex structures based on non-covalent molecular interactions, such as hydrogen bonds or metal coordination. These structures can have innovative applications in fields such as new material design, catalysis, and controlled drug release. However, predicting the products of supramolecular reactions is particularly complex, as interactions between molecules are often difficult to control. In this application, the system combined three different amines, three pyridines containing carbonyl groups, and metals such as Cu+ and Zn2+, in an attempt to form supramolecular structures. The reactions were conducted in parallel using small reaction volumes, between 2 and 4 mL, and the obtained samples were analyzed by direct mass spectrometry and NMR spectroscopy to identify the most promising structures.
The decision algorithm evaluated each reaction based on the presence of specific patterns in the NMR and MS data. In total, 54 different experiments were conducted to explore the vast combination of reagents and reaction conditions. Each reaction was replicated at least six times to ensure the reproducibility and robustness of the results obtained. Direct mass spectrometry was used to determine the presence of metal-organic assemblies in different oxidation states, allowing desired products to be distinguished from reaction intermediates.
Two main supramolecular structures were identified: a metal-organic cage [Zn4(L1)4]8+ and a metal-organic helix [Zn2(L2)3]4+. The metal-organic cage was observed in 12 out of 54 experiments, while the metal-organic helix was obtained in 8 experiments. Both structures were further analyzed to assess their ability to bind molecular guests, using a library of six small organic molecules as guests.
The binding test was conducted under non-coordinating solvent conditions, and each candidate guest was introduced in increasing concentrations until reaching a molar ratio of 1:5 relative to the host structure. The decision algorithm analyzed changes in the NMR spectra as the guest concentration varied, identifying cases where significant chemical shifts occurred. The metal-organic cage showed a strong affinity for three of the six guest molecules, suggesting the formation of stable inclusion complexes, while the helix showed no significant binding with any of the guests.
The autonomous supramolecular synthesis and characterization process was conducted for three consecutive days without human intervention, leading to the analysis of over 300 samples and the generation of more than 7 GB of data. These data were used not only to identify the formed complexes but also to further optimize the synthesis conditions, reducing the number of steps needed to obtain pure and stable structures. The use of techniques such as low-resolution NMR spectroscopy and direct mass spectrometry allowed for rapid and reliable screening of the properties of the supramolecular complexes.
This approach demonstrated the extension of autonomy to the evaluation of the functional properties of compounds, with potential applications in catalysis and advanced materials design.
Offline Photocatalytic Synthesis
The autonomous platform is designed to be easily expandable. In a demonstration experiment, a commercial photoreactor was added as an external module to perform a photocatalytic synthesis of a carboxylic acid derivative. The samples were transported by mobile robots to the photoreactor, SynLED, for irradiation with light at 465 nm for a time ranging from 2 to 6 hours, depending on the optimal reaction conditions. The photoreactor has a capacity of 16 vials simultaneously, allowing reactions to be carried out in parallel.
The samples were irradiated with a power density of 25 mW/cm², and during the reaction, temperatures were maintained between 20°C and 30°C using an integrated cooling system. At the end of the photocatalytic process, the samples were automatically transferred by mobile robots to the ISynth platform for subsequent analysis and purification phases.
The analysis of the irradiated samples was conducted with UPLC-MS to identify the reaction products. Chromatographic separation was performed on a UHPLC BEH C18 column with a mobile phase consisting of acetonitrile and water (with 0.1% formic acid) in a gradient from 5% to 95% acetonitrile in 3 minutes, followed by a 1-minute re-equilibration period. Product detection showed a conversion efficiency of 65-85%, depending on the type of catalyst used.
Several photocatalysts were tested, including [Ir(dtbbpy)(ppy)2]PF6, eosin Y, and 4CzIPN. The catalyst [Ir(dtbbpy)(ppy)2]PF6 showed the best performance with an 85% conversion efficiency and high selectivity towards the desired product. Eosin Y, on the other hand, showed lower efficiency, reaching only 65% conversion. Additionally, the autonomous system was able to monitor the reproducibility of the results by performing three replicates for each experiment and analyzing any deviations in conversion and selectivity profiles.
The entire photocatalytic process lasted about five days, during which over 150 samples were processed. The system generated and analyzed more than 10 GB of data, using these results to further refine experimental parameters, such as irradiation duration and photocatalyst concentration. The data indicated that optimizing reaction conditions could reduce energy consumption by up to 20% while maintaining high product yield.
Future Perspectives
This study demonstrates that the use of mobile robots to automate chemical synthesis and analysis is feasible and that such systems can emulate human chemists' decisions in exploratory contexts. Although the level of contextual understanding is still lower than that of humans, the ability to make algorithmic decisions instantly offers a significant speed advantage.
For the future, one of the main objectives will be the integration of advanced artificial intelligence algorithms, such as deep neural networks, to further improve the system's decision-making capabilities. These algorithms could be trained on large amounts of data collected from the laboratory, allowing for greater accuracy in recognizing complex patterns and better handling of reaction conditions. The introduction of advanced machine learning algorithms could increase the system's ability to identify promising reactions with a success rate 20% higher than current heuristic-based methods.
Additionally, the adoption of predictive models for selecting optimal experimental conditions could reduce reaction optimization times by up to 30%, enabling faster discovery of new compounds. These models, based on methods such as Bayesian optimization, could guide the platform towards choosing the best combinations of reagents, temperatures, and catalysts, drastically reducing the number of experiments needed.
Another area of development is increasing automation in reagent and consumable management. Currently, reagent replacement requires human intervention, but the future introduction of automated storage systems and material handling robots could allow for even greater autonomy. It is expected that a fully automated system could reduce downtime related to replenishment by up to 40%, further improving laboratory efficiency.
The integration of advanced sensors for in situ monitoring of reactions could provide real-time data on parameters such as pH, viscosity, and temperature with a precision of ±0.1%. These sensors, combined with adaptive control algorithms, could allow for automatic adjustment of reaction conditions, improving the yield and quality of the final product. It is estimated that the use of such technologies could increase overall reaction yields by 10-15% compared to current methods.
Finally, an important direction will be expanding synthesis capabilities at the industrial level. While the current platform is designed for research laboratories, implementing robotic systems on a larger scale, with the ability to handle reaction volumes of up to 100 L, could transform large-scale chemical production. This could reduce production costs by 25-30% and improve operational safety by reducing human intervention in potentially hazardous processes.
Conclusions
The integration of autonomous robotics and exploratory chemistry represents one of the most exciting frontiers for the future of chemical research and production. Advanced automation, such as that described in this study, highlights how shifting from a reactive to a proactive approach in chemical discovery can have significant implications. Indeed, while traditional laboratories are limited to following standard procedures, these autonomous systems can "learn" from intermediate results and adapt in real-time, emulating human intuition and reasoning but operating on a much larger and faster scale. This not only accelerates the discovery process but also paves the way for generating new molecules and chemical structures that are impossible to predict in advance.
This technology, in addition to reducing development time, also creates a cultural shift in how scientific experimentation is conceived. In an autonomous environment, the added value of scientists no longer lies in performing manual operations but in designing decision-making strategies and algorithms that can fully exploit the robots' capabilities. This paradigm shift underscores how artificial intelligence is transforming the human role into an "orchestrator" of decision-making processes, with a significant impact not only for chemistry but also for all scientific and industrial disciplines that require the exploration of large volumes of possibilities.
For companies, this represents a real competitive transformation. The ability to scale autonomous chemical synthesis up to industrial production would entail cutting production costs, reducing operational risks, and offering unprecedented flexibility in introducing new molecules and materials to the market in reduced times. The modularity of the system allows companies to quickly expand or modify their laboratories' capabilities according to needs, ensuring resilience that becomes a strategic advantage in adapting to emerging trends and demands.
Furthermore, the ability to acquire and process data accurately allows for an in-depth understanding of chemical properties and production processes, which is essential for improving the environmental sustainability of industrial chemistry. Reducing waste and optimizing energy consumption, made possible by automatic reaction condition regulation, aligns exploratory chemistry with increasingly stringent sustainability policies, transforming this technology into a driver for responsible innovation.
In the future, the evolution of predictive models and automation of material management will make these systems not only increasingly efficient but also capable of supporting large-scale customization of chemical products, paving the way for the creation of "tailor-made" molecules for sectors such as pharmaceuticals, advanced materials, and energy. Companies investing in these technologies will not only gain an immediate competitive advantage but, ultimately, redefine the boundaries of what can be achieved in the chemical-industrial field, projecting the sector towards a model of continuous and autonomous innovation.
Commentaires