Beyond the Numbers: Charting Chemical Reaction Space

Leveraging Limited Experimental Data with Machine Learning: Differentiating a Methyl from an Ethyl Group in the Corey-Bakshi-Shibata Reduction

We present a case study on how to improve an existing metal-free catalyst for a particularly difficult reaction, namely, the Corey-Bakshi-Shibata (CBS) reduction of butanone, which constitutes the classic and prototypical challenge of being able to differentiate a methyl from an ethyl group. As there are no known strategies on how to address this challenge, we leveraged the power of machine learning by constructing a realistic (for a typical laboratory) small, albeit high-quality, data set of about 100 reactions (run in triplicate) that we used to train a model in combination with a key-intermediate graph (of substrate and catalyst) to predict the differences in Gibbs activation energies ΔΔG‡ of the enantiomeric reaction paths. With the help of this model, we were able to select and subsequently screen a small selection of catalysts and increase the selectivity for the CBS reduction of butanone to 80% enantiomeric excess (ee), the highest possible value achieved to date for this substrate with a metal-free catalyst, thereby also exceeding the best available enzymatic systems (64% ee) and the selectivity with Corey's original catalyst (60% ee). This translates into a >50% improvement in relative ΔG‡ from 0.9 to 1.4 kcal mol-1. We underscore the transformative potential of machine learning in accelerating catalyst design because we rely on a manageable small data set and a key-intermediate graph representing a combination of catalyst and substrate graphs in lieu of a transition-state model. Our results highlight the synergy of synthetic chemistry and data-centric approaches and provide a blueprint for future catalyst optimization.

Inside the Borate Anomaly: Leveraging a Predictive Modelling Approach to Navigate Complex Composition-Structure-Property Relationships in Oxyhalide Borate Glasses

This study employs a systematic and predictive modelling approach to investigate the structure and properties of multi-component borate glasses. In particular, this work is focused on understanding the individual and interaction effects of multiple constituents on several material properties. By leveraging advanced modeling techniques, this work examines how the inclusion and variation of B2O3, CaF2, TiO2, ZnO, and Na2CO3 influence the glass network, with particular attention to modifier fractions ≥ 30 mol%. This research addresses the gap in knowledge regarding the complex behavior of borate glasses in this high modifier fraction range, known as the borate anomaly, where prediction of glass structure and properties becomes particularly challenging. The use of a design of mixtures (DoM) approach facilitated the generation of polynomial equations indicating the influence of mixture components on various responses, enabling the prediction and optimization of glass properties over broad compositional ranges despite being within the anomalous region. This methodical approach not only advances our understanding of borate glass systems but also underscores the importance of predictive modelling in the accelerated design and development of novel glass materials for diverse applications.

When Yield Prediction Does Not Yield Prediction: An Overview of the Current Challenges

Machine Learning (ML) techniques face significant challenges when predicting advanced chemical properties, such as yield, feasibility of chemical synthesis, and optimal reaction conditions. These challenges stem from the high-dimensional nature of the prediction task and the myriad essential variables involved, ranging from reactants and reagents to catalysts, temperature, and purification processes. Successfully developing a reliable predictive model not only holds the potential for optimizing high-throughput experiments but can also elevate existing retrosynthetic predictive approaches and bolster a plethora of applications within the field. In this review, we systematically evaluate the efficacy of current ML methodologies in chemoinformatics, shedding light on their milestones and inherent limitations. Additionally, a detailed examination of a representative case study provides insights into the prevailing issues related to data availability and transferability in the discipline.

Charting the Chemical Reaction Space around a Multicomponent Combination: Controlled Access to a Diverse Set of Biologically Relevant Scaffolds

Charting the chemical reaction space around the combination of carbonyls, amines, and isocyanoacetates allows the description of new multicomponent processes leading to a variety of unsaturated imidazolone scaffolds. The resulting compounds display the chromophore of the green fluorescent protein and the core of the natural product coelenterazine. Despite the competitive nature of the pathways involved, general protocols provide selective access to the desired chemotypes. Moreover, we describe unprecedented reactivity at the C-2 position of the imidazolone core to directly afford C, S, and N-derivatives featuring natural products (e.g. leucettamines), potent kinase inhibitors, and fluorescent probes with suitable optical and biological profiles.

Charting the Chemical Reaction Space around a Multicomponent Combination: Controlled Access to a Diverse Set of Biologically Relevant Scaffolds

Charting the chemical reaction space around the combination of carbonyls, amines, and isocyanoacetates allows the description of new multicomponent processes leading to a variety of unsaturated imidazolone scaffolds. The resulting compounds display the chromophore of the green fluorescent protein and the core of the natural product coelenterazine. Despite the competitive nature of the pathways involved, general protocols provide selective access to the desired chemotypes. Moreover, we describe unprecedented reactivity at the C-2 position of the imidazolone core to directly afford C, S, and N-derivatives featuring natural products (e.g. leucettamines), potent kinase inhibitors, and fluorescent probes with suitable optical and biological profiles.

Probing the Impact of Solvent on the Strength of Lewis Acids via Fluorescent Lewis Adducts

Various methods have been developed to measure the strength of a Lewis acid. A major challenge for these measurements lies in the complexity that arises from variable solvent interactions and perturbations of Lewis acids as their reaction environment changes. Herein, we investigate the impact of solvent effects on Lewis acids for the first time as measured by the fluorescent Lewis adduct (FLA) method. The binding of a Lewis acid in various solvents reveals a measurable dichotomy between both polarity and donor ability of the solvent. While not strictly separable, we observe that the influence of solvent polarity on Lewis acid unit (LAU) values is distinctly opposite to the influence of donor ability. This dichotomy was confirmed by titration data, illustrating that solvation effects can be appropriately and precisely gauged by the FLA method.

A Review on Artificial Intelligence Enabled Design, Synthesis, and Process Optimization of Chemical Products for Industry 4.0

With the development of Industry 4.0, artificial intelligence (AI) is gaining increasing attention for its performance in solving particularly complex problems in industrial chemistry and chemical engineering. Therefore, this review provides an overview of the application of AI techniques, in particular machine learning, in chemical design, synthesis, and process optimization over the past years. In this review, the focus is on the application of AI for structure-function relationship analysis, synthetic route planning, and automated synthesis. Finally, we discuss the challenges and future of AI in making chemical products.

Using Active Learning to Develop Machine Learning Models for Reaction Yield Prediction

Computer aided synthesis planning, suggesting synthetic routes for molecules of interest, is a rapidly growing field. The machine learning methods used are often dependent on access to large datasets for training, but finite experimental budgets limit how much data can be obtained from experiments. This suggests the use of schemes for data collection such as active learning, which identifies the data points of highest impact for model accuracy, and which has been used in recent studies with success. However, little has been done to explore the robustness of the methods predicting reaction yield when used together with active learning to reduce the amount of experimental data needed for training. This study aims to investigate the influence of machine learning algorithms and the number of initial data points on reaction yield prediction for two public high-throughput experimentation datasets. Our results show that active learning based on output margin reached a pre-defined AUROC faster than random sampling on both datasets. Analysis of feature importance of the trained machine learning models suggests active learning had a larger influence on the model accuracy when only a few features were important for the model prediction.

Computational Mapping of Dirhodium(II) Catalysts

The chemistry of dirhodium(II) catalysts is highly diverse, and can enable the synthesis of many different molecular classes. A tool to aid in catalyst selection, independent of mechanism and reactivity, would therefore be highly desirable. Here, we describe the development of a database for dirhodium(II) catalysts that is based on the principal component analysis of DFT-calculated parameters capturing their steric and electronic properties. This database maps the relevant catalyst space, and may facilitate exploration of the reactivity landscape for any process catalysed by dirhodium(II) complexes. We have shown that one of the principal components of these catalysts correlates with the outcome (e.g. yield, selectivity) of a transformation used in a molecular discovery project. Furthermore, we envisage that this approach will assist the selection of more effective catalyst screening sets, and, hence, the data-led optimisation of a wide range of rhodium-catalysed transformations.

The AI for Scientific Discovery Network. PATTERNS (NEW YORK, N.Y.) 2021;2:100162. [PMID: 33511363 PMCID: PMC7815949 DOI: 10.1016/j.patter.2020.100162]

The Artificial Intelligence and Augmented Intelligence for Automated Investigation for Scientific Discovery Network+ (AI3SD) was established in response to the UK Engineering and Physical Sciences Research Council (EPSRC) late-2017 call for a Network+ to promote cutting-edge research in artificial intelligence to accelerate groundbreaking scientific discoveries. This article provides the philosophical, scientific, and technical underpinnings of the Network+, the history of the different domains represented in the Network+, and the specific focus of the Network+. The activities, collaborations, and research covered in the first year of the Network+ have highlighted the significant challenges in the chemistry and augmented and artificial intelligence space. These challenges are shaping the future directions of the Network+. The article concludes with a summary of the lessons learned in running this Network+ and introduces our plans for the future in a landscape redrawn by COVID-19, including rebranding into the AI 4 Scientific Discovery Network (www.ai4science.network).

Achieving selectivity in porphyrin bromination through a DoE-driven optimization under continuous flow conditions

The post-functionalization of porphyrins through the bromination in β position of the pyrrolic rings is a relevant transformation because the resulting bromoderivatives are useful synthons to covalently link a variety of chemical architectures to a porphyrin ring. However, single bromination of porphyrins is a challenging reaction for the abundancy of reactive β-pyrrolic positions in the aromatic macrocycle. We herein report a synthetic procedure for the efficient preparation of 2-bromo-5,10,15,20-tetraphenylporphyrin (1) under continuous flow conditions. The use of flow technology allows to reach an accurate control over critical reaction parameters such as temperature and reaction time. Furthermore, by performing the optimization process through a statistical DoE (Design of Experiment) approach, these parameters could be properly adjusted with a limited number of experiments. This process led us to a better understanding of the relevant factors that govern porphyrins monobromination and to obtain compound 1 with an unprecedent 80% yield.

Automated generation of photochemical reaction data by transient flow experiments coupled with online HPLC analysis

Competing homo- and crossdimerization reactions between coumarin and 1-methyl-2-quinolinone are investigated by transient continuous-flow experiments combined with online HPLC, enabling the generation and acquisition of large reaction data sets.

Iterative experimental design based on active machine learning reduces the experimental burden associated with reaction screening

Through iterative selection of maximally informative experiments, active learning renders exhaustive screening obsolete. Chosen experiments are used to train models that are accurate over the entire domain, thus reducing the experiment burden.

A Design of Experiments (DoE) Approach Accelerates the Optimization of Copper-Mediated 18F-Fluorination Reactions of Arylstannanes

Recent advancements in ¹⁸F radiochemistry, such as the advent of copper-mediated radiofluorination (CMRF) chemistry, have provided unprecedented access to novel chemically diverse PET probes; however, these multicomponent reactions have come with a new set of complex optimization problems. Design of experiments (DoE) is a statistical approach to process optimization that is used across a variety of industries. It possesses a number of advantages over the traditionally employed "one variable at a time" (OVAT) approach, such as increased experimental efficiency as well as an ability to resolve factor interactions and provide detailed maps of a process's behavior. Here we demonstrate the utility of DoE to the development and optimization of new radiochemical methodologies and novel PET tracer synthesis. Using DoE to construct experimentally efficient factor screening and optimization studies, we were able to identify critical factors and model their behavior with more than two-fold greater experimental efficiency than the traditional OVAT approach. Additionally, the use of DoE allowed us to glean new insights into the behavior of the CMRF of a number of arylstannane precursors. This information has guided our decision-making efforts while developing efficient reaction conditions that suit the unique process requirements of ¹⁸F PET tracer synthesis.

Machine learning and molecular descriptors enable rational solvent selection in asymmetric catalysis

Rational solvent selection remains a significant challenge in process development. Here we describe a hybrid mechanistic-machine learning approach, geared towards automated process development workflow. A library of 459 solvents was used, for which 12 conventional molecular descriptors, two reaction-specific descriptors, and additional descriptors based on screening charge density, were calculated. Gaussian process surrogate models were trained on experimental data from a Rh(CO)2(acac)/Josiphos catalysed asymmetric hydrogenation of a chiral α-β unsaturated γ-lactam. With two simultaneous objectives - high conversion and high diastereomeric excess - the multi-objective algorithm, trained on the initial dataset of 25 solvents, has identified solvents leading to better reaction outcomes. In addition to being a powerful design of experiments (DoE) methodology, the resulting Gaussian process surrogate model for conversion is, in statistical terms, predictive, with a cross-validation correlation coefficient of 0.84. After identifying promising solvents, the composition of solvent mixtures and optimal reaction temperature were found using a black-box Bayesian optimisation. We then demonstrated the application of a new genetic programming approach to select an appropriate machine learning model for a specific physical system, which should allow the transition of the overall process development workflow into the future robotic laboratories.

Design of enzymatic cascade processes for the production of low-priced chemicals

While the application of enzymes to synthetic and industrial problems continues to grow, the major development today is focused on multi-enzymatic cascades. Such systems are particularly attractive, because many commercially available enzymes operate under relatively similar operating conditions. This opens the possibility of one-pot operation with multiple enzymes in a single reactor. In this paper the concept of modules is introduced whereby groups of enzymes are combined in modules, each operating in a single reactor, but with the option of various operating strategies to avoid any complications of nonproductive interactions between the enzymes, substrates or products in a given reactor. In this paper the selection of modules is illustrated using the synthesis of the bulk chemical, gluconic acid, from lignocellulosic waste.

Computational Ligand Descriptors for Catalyst Design

Ligands, especially phosphines and carbenes, can play a key role in modifying and controlling homogeneous organometallic catalysts, and they often provide a convenient approach to fine-tuning the performance of known catalysts. The measurable outcomes of such catalyst modifications (yields, rates, selectivity) can be set into context by establishing their relationship to steric and electronic descriptors of ligand properties, and such models can guide the discovery, optimization, and design of catalysts. In this review we present a survey of calculated ligand descriptors, with a particular focus on homogeneous organometallic catalysis. A range of different approaches to calculating steric and electronic parameters are set out and compared, and we have collected descriptors for a range of representative ligand sets, including 30 monodentate phosphorus(III) donor ligands, 23 bidentate P,P-donor ligands, and 30 carbenes, with a view to providing a useful resource for analysis to practitioners. In addition, several case studies of applications of such descriptors, covering both maps and models, have been reviewed, illustrating how descriptor-led studies of catalysis can inform experiments and highlighting good practice for model comparison and evaluation.

High throughput reaction screening using desorption electrospray ionization mass spectrometry

We report the high throughput analysis of reaction mixture arrays using methods and data handling routines that were originally developed for biological tissue imaging.

We report the high throughput analysis of reaction mixture arrays using methods and data handling routines that were originally developed for biological tissue imaging. Desorption electrospray ionization (DESI) mass spectrometry (MS) is applied in a continuous on-line process at rates that approach 10⁴ reactions per h at area densities of up to 1 spot per mm² (6144 spots per standard microtiter plate) with the sprayer moving at ca. 10⁴ microns per s. Data are analyzed automatically by MS using in-house software to create ion images of selected reagents and products as intensity plots in standard array format. Amine alkylation reactions were used to optimize the system performance on PTFE membrane substrates using methanol as the DESI spray/analysis solvent. Reaction times can be <100 μs when reaction acceleration occurs in microdroplets, enabling the rapid screening of processes like N-alkylation and Suzuki coupling reactions as reported herein. Products and by-products were confirmed by on-line MS/MS upon rescanning of the array.

Optimum catalyst selection over continuous and discrete process variables with a single droplet microfluidic reaction platform

A mixed-integer nonlinear program (MINLP) algorithm to optimize catalyst turnover number (TON) and product yield by simultaneously modulating discrete variables—catalyst types—and continuous variables—temperature, residence time, and catalyst loading—was implemented and validated.

Exploring experimental fitness landscapes for chemical synthesis and property optimization

The topology of experimental fitness landscapes for chemical optimization objectives is assessed through svr-based HDMR modeling.

Rationalization of liquid assisted grinding intercalation yields of organic molecules into layered double hydroxides by multivariate analysis

PCA, coupled to molecular descriptors, proved to be an effective tool to rationalize the mechanochemical intercalation yields of layered materials.

Iridium-catalysed ortho-H/D and -H/T exchange under basic conditions: C–H activation of unprotected tetrazoles

Selective C–H activation with unprotected 2-aryltetrazoles has been achieved by base-assisted iridium(i) catalysis to deliver ortho-deuterated and -tritiated tetrazoles.

The application of design of experiments (DoE) reaction optimisation and solvent selection in the development of new synthetic chemistry

This article outlines the benefits of using 'Design of Experiments' (DoE) optimisation during the development of new synthetic methodology. A particularly important factor in the development of new chemical reactions is the choice of solvent which can often drastically alter the efficiency and selectivity of a process. Whilst solvent optimisation is usually done in a non-systematic way based upon a chemist's intuition and previous laboratory experience, we illustrate how optimisation of the solvent for a reaction can be carried out by using a 'map of solvent space' in a DoE optimisation. A new solvent map has been developed specifically for optimisation of new chemical reactions using principle component analysis (PCA) incorporating 136 solvents with a wide range of properties. The new solvent map has been used to identify safer alternatives to toxic/hazardous solvents, and also in the optimisation of an S(N)Ar reaction.

B(OCH2CF3)3-mediated direct amidation of pharmaceutically relevant building blocks in cyclopentyl methyl ether

The use of B(OCH2CF3)3 for mediating direct amidation reactions of a wide range of pharmaceutically relevant carboxylic acids and amines is described, including numerous heterocycle-containing examples. An initial screen of solvents for the direct amidation reaction suggested that cyclopentyl methyl ether, a solvent with a very good safety profile suitable for use over a wide temperature range, was an excellent replacement for the previously used solvent acetonitrile. Under these conditions amides could be prepared from 18 of the 21 carboxylic acids and 18 of the 21 amines examined. Further optimisation of one of the low yielding amidation reactions (36% yield) via a design of experiments approach enabled an 84% yield of the amide to be obtained.

Nanomole-scale high-throughput chemistry for the synthesis of complex molecules

At the forefront of new synthetic endeavors, such as drug discovery or natural product synthesis, large quantities of material are rarely available and timelines are tight. A miniaturized automation platform enabling high-throughput experimentation for synthetic route scouting to identify conditions for preparative reaction scale-up would be a transformative advance. Because automated, miniaturized chemistry is difficult to carry out in the presence of solids or volatile organic solvents, most of the synthetic “toolkit” cannot be readily miniaturized. Using palladium-catalyzed cross-coupling reactions as a test case, we developed automation-friendly reactions to run in dimethyl sulfoxide at room temperature. This advance enabled us to couple the robotics used in biotechnology with emerging mass spectrometry–based high-throughput analysis techniques. More than 1500 chemistry experiments were carried out in less than a day, using as little as 0.02 milligrams of material per reaction.