A Brief Introduction to Chemical Reaction Optimization

Abstract

From the start of a synthetic chemist's training, experiments are conducted based on recipes from textbooks and manuscripts that achieve clean reaction outcomes, allowing the scientist to develop practical skills and some chemical intuition. This procedure is often kept long into a researcher's career, as new recipes are developed based on similar reaction protocols, and intuition-guided deviations are conducted through learning from failed experiments. However, when attempting to understand chemical systems of interest, it has been shown that model-based, algorithm-based, and miniaturized high-throughput techniques outperform human chemical intuition and achieve reaction optimization in a much more time- and material-efficient manner; this is covered in detail in this paper. As many synthetic chemists are not exposed to these techniques in undergraduate teaching, this leads to a disproportionate number of scientists that wish to optimize their reactions but are unable to use these methodologies or are simply unaware of their existence. This review highlights the basics, and the cutting-edge, of modern chemical reaction optimization as well as its relation to process scale-up and can thereby serve as a reference for inspired scientists for each of these techniques, detailing several of their respective applications.

Figure 1

An example of an OFAT experimental procedure in varying temperature and reagent equivalents, where ○ represents a numbered experimental data point and the blue region indicates the true optimum area of parameter space. Response surface is contoured from red (low response) to blue (high response).

Scheme 1. Model Reaction Used for OFAT Optimization for the Synthesis of Propargylamine Derivatives

Scheme 2. S_NAr System of Interest, Where the DoE Campaign Aims to Optimize the Yield of the Ortho-Substituted Product, 7(1)

Figure 2

Contour plot for the response of 7, showing how the yield of the desired product varies with respect to changing experimental conditions.

Scheme 3. One Reaction of Interest, Optimizing the Yield and Selectivity of the Desired 3,4-Dihydroxymandelic Acid Intermediate, 12

This intermediate can then be used to synthesize either vanillin, 13, iso-vanillin, 14, or heliotropin, 15.

Figure 3

Contour plot for the selectivity of the reaction forming the desired intermediate, 12. Data was used from the original publication by Minisci and co-workers to refit the model and plot the response surface using MODDE Pro.

Figure 4

Parameter space exploration expected when comparing a typical OFAT optimization with a DoE design, where • represents an experiment. The DoE shown represents a CCF experimental design. Note that an OFAT optimization does not require a predetermined number of experiments and therefore may or may not exceed the number of experiments in a given DoE design.

Scheme 4. Alkylation of the Indolphenol, 16, with the Chloropyrrolidine, 17, to Form the Desired Cediranib Product, 18

This reaction was found to proceed via the azetidinium intermediate, 19, as a result of kinetic modelling.

Scheme 5. Aqueous Reduction of 4-Nitrophenol, 20, to 4-Aminophenol, 21, Using Gold Nanoparticles (AuNPs) and NaBH₄

Figure 5

Kinetics-derived response surface for the conversion of 4-nitrophenol, 20, to 4-aminophenol, 21, when exploring the variables of residence time and AuNP surface area per liter.

Figure 6

Common conventional kinetic analysis techniques for the determination of: (a) unimolecular zero-order kinetics, (b) unimolecular first-order kinetics, (c) bimolecular second-order kinetics between the same reactants, and (d) bimolecular second-order kinetics between different reactants.

Scheme 6. Reaction of Al-Me, 22, with PfBr, 23, to Form the Protected Amino Acid Pf-Al-Me, 24(92)

Figure 7

Kinetic profiles for three kinetic experiments at 30 °C, 35 and 40 °C, where red plots indicate PfBr concentrations and blue plots indicate Pf-Al-Me concentrations. At 30 °C: blue solid squares = experimental data, — = kinetic fit. At 35 °C: blue solid triangles = experimental data, - - - = kinetic fit. At 40 °C: blue solid circles = experimental data, _······ = kinetic fit.

Scheme 7. Stereoselective Reduction of 2-Phenylquinoline, 25, to Yield the Tetrahydroquinoline, 28, Using the Hantzsch Ester, 26, and the Macrocyclic Catalyst, 27(131)

Scheme 8. Cobalt-Catalyzed C–H Functionalization/Alkyne Annulation Reaction of 29 with 30 to Form the Dihydroisoquinoline Product, 31(139)

Figure 8

Three parts of a self-optimizing reactor are an automated reactor system, an analytical method, and an optimization algorithm.

Figure 9

Examples of automated reactors. (a) A bespoke automated flow reactor equipped with pumps, reactors, and analytical equipment. (b) A commercial robotic liquid handler that can be utilized as an automated batch reactor (see section 6.2 for more details).

Figure 10

Development of a Buchwald C–N cross coupling between p-tolyl triflate (32) and aniline (33) via self-optimization in an automated droplet flow reactor. Three continuous variables (residence time, base equivalents, and temperature) and two categorical variables (catalyst and base) were varied to maximize the yield of 4-(p-tolyl)morpholine (34). In the chart, each column contains data for a different catalyst and base combination, and the experiments in each column are shown left to right in the order they were selected by the optimization algorithm. Additionally, the color bar shows experiment selection order.

Figure 11

Development of a stereoselective Suzuki coupling between sulfonate 35-E and boronic acid 36 to form 37-E and 37-Z via self-optimization in an automated batch reactor. In 161 experiments, the yield of 37-E was improved from 30% nominal to 70%, and the E/Z ratio from 1.5:1 to 2.5:1. AY represents assay yield.

Figure 12

Example of a local optimization algorithm failing to find the global maximum of a function with multiple local optima, i.e., the algorithm finds a maximum peak but not the highest peak. X and Y are hypothetical experimental variables (e.g., temperature, reaction time), and the objective is the value that must be maximized (e.g., yield). Unfilled circles are function evaluations (i.e., experiments). Red indicates local maxima function value, while blue indicates local minima.

Figure 13

Multiobjective self-optimization of the N-benzylation of N-benzylation of α-methylbenzylamine 38 with benzyl bromide 39. TSEMO was used to maximize space–time yield of the desired 2° amine 40 and minimize production of the percent impurity of 3° amine 41. After 20 experiments designed by Latin hypercube sampling (LHS), TSEMO quickly identified experiments on or near the Pareto front.

Figure 14

Schematic of a typical HTE workflow where a particular chemical process is optimized with respect to 12 catalysts, 4 bases, and 2 solvents.

Figure 15

Routine HPLC-MS kit equipped with a single-quadruple MS can be used for analysis of ultraHTE reactions direct from 384-well MTP. Two techniques can be employed, either the multiple injections in a single experimental run (MISER) method or a more traditional LCMS/UV method, however, there is a trade-off between speed and the level of quantification that can be achieved.

Figure 16

Platform for automated nanomole-scale reaction screening and micromole-scale synthesis in flow developed by Pfizer and reported for the optimization of a Suzuki reaction.

Figure 17

Schematic of the data collection and model training steps of using machine learning for reaction optimization from public or private reaction databases.

Figure 18

Illustration of the overlap of chemical reaction databases (Reaxys, Pistachio, USPTO, and a subset of AstraZeneca ELN13).

Figure 19

Demonstration of the architecture used by Jensen and co-workers for predicting reaction conditions.

Figure 20

Laboratory vs industrial scale: comparison of characteristic times (in seconds) for mixing, heat transfer and liquid space time observed in reactors used in benchtop optimization and large-scale industrial reactors.^,