Discovering New Chemistry with an Autonomous Robotic Platform Driven by a Reactivity-Seeking Neural Network

Self-Driving Laboratories for Chemistry and Materials Science

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.

Improved on-line benchtop 31P NMR reaction monitoring via Multi-Resonance SHARPER

On-line reaction monitoring of hydrogenation reactions featuring oxygen-sensitive organometallic complexes is done via a 31P benchtop NMR spectrometer using the Multi-Resonance Sensitive Homogeneous And Resolved PEaks in Real time (MR-SHARPER) sequence. Signal enhancement generated by MR-SHARPER enables monitoring of reactivity on the order of minutes that could not be followed with traditional 31P{1H} NMR detection.

Multi-armed bandit algorithm for sequential experiments of molecular properties with dynamic feature selection

Sequential optimization is one of the promising approaches in identifying the optimal candidate(s) (molecules, reactants, drugs, etc.) with desired properties (reaction yield, selectivity, efficacy, etc.) from a large set of potential candidates, while minimizing the number of experiments required. However, the high dimensionality of the feature space (e.g., molecular descriptors) makes it often difficult to utilize the relevant features during the process of updating the set of candidates to be examined. In this article, we developed a new sequential optimization algorithm for molecular problems based on reinforcement learning, multi-armed linear bandit framework, and online, dynamic feature selections in which relevant molecular descriptors are updated along with the experiments. We also designed a stopping condition aimed to guarantee the reliability of the chosen candidate from the dataset pool. The developed algorithm was examined by comparing with Bayesian optimization (BO), using two synthetic datasets and two real datasets in which one dataset includes hydration free energy of molecules and another one includes a free energy difference between enantiomer products in chemical reaction. We found that the dynamic feature selection in representing the desired properties along the experiments provides a better performance (e.g., time required to find the best candidate and stop the experiment) as the overall trend and that our multi-armed linear bandit approach with a dynamic feature selection scheme outperforms the standard BO with fixed feature variables. The comparison of our algorithm to BO with dynamic feature selection is also addressed.

A robotic system for automated chemical synthesis of therapeutic agents

The development of novel compounds for tissue-specific targeting and imaging is often impeded by a lack of lead compounds and the availability of reliable chemistry. Automated chemical synthesis systems provide a potential solution by enabling reliable, repeated access to large compound libraries for screening. Here we report an integrated solid-phase combinatorial chemistry system created using commercial and customized robots. Our goal is to optimize reaction parameters, such as varying temperature, shaking, microwave irradiation, aspirating and dispensing large-sized solid beads, and handling different washing solvents for separation and purification. This automated system accommodates diverse chemical reactions such as peptide synthesis and conventional coupling reactions. To confirm its functionality and reproducibility, 20 nerve-specific contrast agents for biomedical imaging were systematically and repeatedly synthesized and compared to other nerve-targeted agents using molecular fingerprinting and Uniform Manifold Approximation and Projection, which lays the foundation for creating reliable and reproductive chemical libraries in bioimaging and nanomedicine.

Reaction Kinetics using a Chemputable Framework for Data Collection and Analysis

Automated chemistry platforms have been widely explored, but many focus on fixed tasks for chemical synthesis or analysis. However, a typical synthetic chemistry workflow utilizes both, such as kinetic measurements for reaction development and optimization. Due to their repetitive and time-consuming nature, kinetic measurements are often omitted, which limits the mechanistic investigation of reactions. Herein, we present a "Chemputer" platform with on-line analytics (UV/Vis, NMR) which automates routine kinetic measurements. The system's capabilities are showcased by exploring an inverse electron-demand Diels-Alder using initial rate measurements, a metal complexation using variable time normalization analysis (VTNA), and formation of a series of tosylamide derivatives using Hammett analysis. Over 60 individual experiments are presented which required minimal intervention, highlighting the significant time savings of automation. Owing to the modular design of the platform, which facilitates rapid integration of commercial analytical tools, our approach is widely accessible and adjustable to the reaction under investigation. The platform is operated using the chemical programming language, XDL, hence experimental procedures and results are stored in a precise, computer-readable format. We propose that widespread adoption of this reporting protocol in the chemical community could build a database of validated kinetic data beneficial for Machine Learning.

Automated Structural Activity Screening of β-Diketonate Assemblies with High-Throughput Ion Mobility-Mass Spectrometry

Embracing complexity in design, metallo-supramolecular self-assembly presents an opportunity for fabricating materials of economic significance. The array of accessible supramolecules is alluring, along with favourable energy requirements. Implementation is hampered by an inability to efficiently characterise complex mixtures. The stoichiometry, size, shape, guest binding properties and reactivity of individual components and combinations thereof are inherently challenging to resolve. A large combinatorial library of four transition metals (Fe, Cu, Ni and Zn), and six β-diketonate ligands at different molar ratios and pH was robotically prepared and directly analysed over multiple timepoints with electrospray ionisation travelling wave ion mobility-mass spectrometry. The dataset was parsed for self-assembling activity without first attempting to structurally assign individual species. Self-assembling systems were readily categorised without manual data-handling, allowing efficient screening of self-assembly activity. This workflow clarifies solution phase supramolecular assembly processes without manual, bottom-up processing. The complex behaviour of the self-assembling systems was reduced to simpler qualities, which could be automatically processed.

Self-driving laboratories to autonomously navigate the protein fitness landscape

Protein engineering has nearly limitless applications across chemistry, energy and medicine, but creating new proteins with improved or novel functions remains slow, labor-intensive and inefficient. Here we present the Self-driving Autonomous Machines for Protein Landscape Exploration (SAMPLE) platform for fully autonomous protein engineering. SAMPLE is driven by an intelligent agent that learns protein sequence-function relationships, designs new proteins and sends designs to a fully automated robotic system that experimentally tests the designed proteins and provides feedback to improve the agent's understanding of the system. We deploy four SAMPLE agents with the goal of engineering glycoside hydrolase enzymes with enhanced thermal tolerance. Despite showing individual differences in their search behavior, all four agents quickly converge on thermostable enzymes. Self-driving laboratories automate and accelerate the scientific discovery process and hold great potential for the fields of protein engineering and synthetic biology.

Autonomous chemical research with large language models

Transformer-based large language models are making significant strides in various fields, such as natural language processing1-5, biology6,7, chemistry8-10 and computer programming11,12. Here, we show the development and capabilities of Coscientist, an artificial intelligence system driven by GPT-4 that autonomously designs, plans and performs complex experiments by incorporating large language models empowered by tools such as internet and documentation search, code execution and experimental automation. Coscientist showcases its potential for accelerating research across six diverse tasks, including the successful reaction optimization of palladium-catalysed cross-couplings, while exhibiting advanced capabilities for (semi-)autonomous experimental design and execution. Our findings demonstrate the versatility, efficacy and explainability of artificial intelligence systems like Coscientist in advancing research.

Exploration and Insights on Topology Adjustment of Giant Heterometallic Cages Featuring Inorganic Skeletons Assisted by Machine Learning

Inorganic molecular cages are emerging multifunctional molecular-based platforms with the unique merits of rigid skeletons and inherited properties from constituent metal ions. However, the sensitive coordination bonds and vast synthetic space have limited their systematic exploration. Herein, two giant cage-like clusters featuring the organic ligand-directed inorganic skeletons of Ni4[La74Ni104(IDA)96(OH)184(C2O4)12(H2O)76]·(NO3)38·(H2O)120 (La74Ni104, 5 × 5 × 3 - C2O4) and [La84Ni132(IDA)108(OH)168(C2O4)24(NO3)12(H2O)116]·(NO3)72·(H2O)296 (La84Ni132, 5 × 5 × 5 - C2O4) were discovered by a high-throughput synthetic search. With the assistance of machine learning analysis of the experimental data, phase diagrams of the two clusters in a four-parameter synthetic space were depicted. The effect of alkali, oxalate, and other parameters on the formation of clusters and the mechanism regulating the size of two n × m × l clusters were elucidated. This work uses high-throughput synthesis and machine learning methods to improve the efficiency of 3d-4f cluster discovery and finds the highest-nuclearity 3d-4f cluster to date by regulating the size of the n × m × l inorganic cages through oxalate ions, which pushes the synthetic methodology study on elusive inorganic giant cages in a significantly systematic way.

Extracting the Synthetic Route of Pd-Based Catalysts in Methanol Steam Reforming from the Scientific Literature

The structured material synthesis route is crucial for chemists in performing experiments and modern applications such as machine learning material design. With the exponential growth of the chemical literature in recent years, manual extraction from the published literature is time-consuming and labor-intensive. This study focuses on developing an automated method for extracting Pd-based catalyst synthesis routes from the chemical literature. First, a paragraph classification model based on regular expressions is employed to identify paragraphs that contain material synthesis processes. The identified paragraphs are verified using machine learning techniques. Second, natural language processing techniques are applied to automatically parse the material synthesis routes from the identified paragraphs, generate regularized flowcharts, and output structured data. Lastly, we utilized the structured data of the synthesis routes to train machine learning models and predict the performance of the materials. The extracted material entities include the product, preparation method, precursor, support, loading, synthesis operation, and operation condition. This method avoids extensive manual data annotation and improves the scientific literature information acquisition efficiency. The accuracy of the 11 material entities exceeds 80%, and the accuracy of the method, support, precursor, drying time, and reduction time exceeds 90%.

In Pursuit of the Exceptional: Research Directions for Machine Learning in Chemical and Materials Science

Exceptional molecules and materials with one or more extraordinary properties are both technologically valuable and fundamentally interesting, because they often involve new physical phenomena or new compositions that defy expectations. Historically, exceptionality has been achieved through serendipity, but recently, machine learning (ML) and automated experimentation have been widely proposed to accelerate target identification and synthesis planning. In this Perspective, we argue that the data-driven methods commonly used today are well-suited for optimization but not for the realization of new exceptional materials or molecules. Finding such outliers should be possible using ML, but only by shifting away from using traditional ML approaches that tweak the composition, crystal structure, or reaction pathway. We highlight case studies of high-Tc oxide superconductors and superhard materials to demonstrate the challenges of ML-guided discovery and discuss the limitations of automation for this task. We then provide six recommendations for the development of ML methods capable of exceptional materials discovery: (i) Avoid the tyranny of the middle and focus on extrema; (ii) When data are limited, qualitative predictions that provide direction are more valuable than interpolative accuracy; (iii) Sample what can be made and how to make it and defer optimization; (iv) Create room (and look) for the unexpected while pursuing your goal; (v) Try to fill-in-the-blanks of input and output space; (vi) Do not confuse human understanding with model interpretability. We conclude with a description of how these recommendations can be integrated into automated discovery workflows, which should enable the discovery of exceptional molecules and materials.

Robotic Modules for the Programmable Chemputation of Molecules and Materials

Before leveraging big data methods like machine learning and artificial intelligence (AI) in chemistry, there is an imperative need for an affordable, universal digitization standard. This mirrors the foundational requisites of the digital revolution, which demanded standard architectures with precise specifications. Recently, we have developed automated platforms tailored for chemical AI-driven exploration, including the synthesis of molecules, materials, nanomaterials, and formulations. Our focus has been on designing and constructing affordable standard hardware and software modules that serve as a blueprint for chemistry digitization across varied fields. Our platforms can be categorized into four types based on their applications: (i) discovery systems for the exploration of chemical space and novel reactivity, (ii) systems for the synthesis and manufacture of fine chemicals, (iii) platforms for formulation discovery and exploration, and (iv) systems for materials discovery and synthesis. We also highlight the convergent evolution of these platforms through shared hardware, firmware, and software alongside the creation of a unique programming language for chemical and material systems. This programming approach is essential for reliable synthesis, designing experiments, discovery, optimization, and establishing new collaboration standards. Furthermore, it is crucial for verifying literature findings, enhancing experimental outcome reliability, and fostering collaboration and sharing of unsuccessful experiments across different research labs.

Machine Learning Strategies for Reaction Development: Toward the Low-Data Limit

Machine learning models are increasingly being utilized to predict outcomes of organic chemical reactions. A large amount of reaction data is used to train these models, which is in stark contrast to how expert chemists discover and develop new reactions by leveraging information from a small number of relevant transformations. Transfer learning and active learning are two strategies that can operate in low-data situations, which may help fill this gap and promote the use of machine learning for tackling real-world challenges in organic synthesis. This Perspective introduces active and transfer learning and connects these to potential opportunities and directions for further research, especially in the area of prospective development of chemical transformations.

Machine learning for yield prediction for chemical reactions using in situ sensors

Machine learning models were developed to predict product formation from time-series reaction data for ten Buchwald-Hartwig coupling reactions. The data was provided by DeepMatter and was collected in their DigitalGlassware cloud platform. The reaction probe has 12 sensors to measure properties of interest, including temperature, pressure, and colour. Colour was a good predictor of product formation for this reaction and machine learning models were able to learn which of the properties were important. Predictions for the current product formation (in terms of % yield) had a mean absolute error of 1.2%. For predicting 30, 60 and 120 min ahead the error rose to 3.4, 4.1 and 4.6%, respectively. The work here presents an example into the insight that can be obtained from applying machine learning methods to sensor data in synthetic chemistry.

Sidekick: A Low-Cost Open-Source 3D-printed liquid dispensing robot

The Sidekick is a desktop liquid dispenser, compatible with standard SBS microplates and designed for accessible laboratory automation. It features an armature-based motion system and a fully 3D-printed chassis to reduce overall mechanical complexity and accommodate user modification. Liquid dispensing is achieved with four commercially available solenoid driven positive displacement pumps that deliver liquid in 10 µL increments. A Raspberry Pi Pico RP2040 processor programmed in MicroPython is used for control, and exposes a USB serial interface for users to submit commands using either a simple vocabulary of commands or a subset of G-Code. At a total cost of $710 USD, the Sidekick offers laboratories an easy to build, easily maintained, open-source liquid dispensing system for both research and pedagogical introductions to lab automation.

Machine-Learning-Guided Identification of Coordination Polymer Ligands for Crystallizing Separation of Cs/Sr

Separation of Cs/Sr is one of many coordination-chemistry-centered processes in the grand scheme of spent nuclear fuel reprocessing, a critical link for a sustainable nuclear energy industry. To deploy a crystallizing Cs/Sr separation technology, we planned to systematically screen and identify candidate ligands that can efficiently and selectively bind to Sr2+ and form coordination polymers. Therefore, we mined the Cambridge Structural Database for characteristic structural information and developed a machine-learning-guided methodology for ligand evaluation. The optimized machine-learning model, correlating the molecular structures of the ligands with the predicted coordinative properties, generated a ranking list of potential compounds for Cs/Sr selective crystallization. The Sr2+ sequestration capability and selectivity over Cs+ of the promising ligands identified (squaric acid and chloranilic acid) were subsequently confirmed experimentally, with commendable performances, corroborating the artificial-intelligence-guided strategy.