Ultrahigh-Throughput Experimentation for Information-Rich Chemical Synthesis

Abstract

Metrics & MoreArticle Recommendations CONSPECTUS:The incorporation of data science is revolutionizing organic chemistry. It is becoming increasingly possible to predict reaction outcomes with accuracy, computationally plan new retrosynthetic routes to complex molecules, and design molecules with sophisticated functions. Critical to these developments has been statistical analysis of reaction data, for instance with machine learning, yet there is very little reaction data available upon which to build models. Reaction data can be mined from the literature, but experimental data tends to be reported in a text format that is difficult for computers to read. Compounding the issue, literature data are heavily biased toward "productive" reactions, and few "negative" reaction data points are reported even though they are critical for training of statistical models. High-throughput experimentation (HTE) has evolved over the past few decades as a tool for experimental reaction development. The beauty of HTE is that reactions are run in a systematic format, so data points are internally consistent, the reaction data are reported whether the desired product is observed or not, and automation may reduce the occurrence of false positive or negative data points. Additionally, experimental workflows for HTE lead to datasets with reaction metadata that are captured in a machine-readable format. We believe that HTE will play an increasingly important role in the data revolution of chemical synthesis. This Account details the miniaturization of synthetic chemistry culminating in ultrahighthroughput experimentation (ultraHTE), wherein reactions are run in ∼1 μL droplets inside of 1536-well microtiter plates to minimize the use of starting materials while maximizing the output of experimental information. The performance of ultraHTE in 1536-well microtiter plates has led to an explosion of available reaction data, which have been used to identify specific substrate− catalyst pairs for maximal efficiency in novel cross-coupling reactions. The first iteration of ultraHTE focused on the use of dimethyl sulfoxide (DMSO) as a high-boiling solvent that is compatible with the plastics most commonly used in consumable well plates, which generated homogeneous reaction mixtures that are perfect for use with nanoliter-dosing liquid handling robotics. In this way, DMSO enabled diverse reagents to be arrayed in ∼1 μL droplets. Reactions were run at room temperature with no agitation and could be scaled up from the ∼0.05 mg reaction scale to the 1 g scale. Engineering enhancements enabled the use of ultraHTE with diverse and semivolatile solvents, photoredox catalysis, heating, and acoustic agitation. A main driver in the development of ultraHTE was the recognition of the opportunity for a direct merger between miniaturized reactions and biochemical assays. Indeed, a strategy was developed to feed ultraHTE reaction mixtures directly to a mass-spectrometry-based affinity selection bioassay. Thus, micrograms of starting materials could be used in t...

Chemputation and the Standardization of Chemical Informatics

“…This combinatorial chemistry and high-throughput screening paradigm was often blamed for a decline in productivity in the pharmaceutical industry, and yet it became a valuable research tool whose relevance is underlined by the commercial availability of mature XYZ Gantry (Chemspeed, Tecan) and ultralow-volume pipetting platforms (Mosquito). In recent years, we have observed a renaissance in the use of massively parallel, miniaturized ultrahigh-throughput experimentation ,− combined with design of experiments (DoE) and other screening techniques applied to these reactors for discovering and optimizing novel reactivity, properties and even bioactivity, though not necessarily isolation procedures . These reactors, usually based on 96-, 384-, or 1536-well plate type designs, and allow hundreds of reactions to be run at once under the same process conditions .…”

A reactivity model for oxidative addition to palladium enables quantitative predictions for catalytic cross-coupling reactions

“…Emerging approaches in reactivity prediction that combine high-throughput experimentation [8][9][10][11][12][13][14] with molecular descriptor sets [15][16][17][18][19][20][21][22][23][24] and multivariate statistical analysis including machine learning [25][26][27][28][29][30][31][32][33][34] can accelerate the screening/optimization process and increase success rates; however, predictions generated by these approaches are oen limited to the specic reaction under investigation. Developing and rening the next generation of organic chemistry tools, including computer-aided synthesis design, automated reaction optimization, and predictive algorithms, 35 requires the development of general and quantitative frameworks that rapidly link molecular structure to reactivity for many different reactants and catalysts.…”

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

“…High-throughput experimentation (HTE) has emerged as a time-and material efficient technique for producing large amounts of chemical reaction data [5,6]. HTE is thus a reasonable approach to yield datasets for CASP, which requires large training sets.…”