How the University of Nottingham streamlined product formation predictions with SmartChemistry® 

The application of artificial intelligence (AI) in chemical reactions is a rapidly growing field, with significant benefits for the chemical industry. Despite these significant advances in automation, most synthetic chemistry is completed manually in glassware as batch chemistry, rather than utilising the applications of machine learning. When properly trained, however, AI can develop intelligent algorithms that analyse large amounts of data to provide quick and accurate predictions for synthetic product formations in chemistry.  

A team of researchers at the University of Nottingham, led by Prof. Jonathan Hirst, recently partnered with deepmatter® to assess how data collected through SmartChemistry® could be analysed and replicated around the benefits of using in-situ sensors to elevate the efficiency of experiments. Read on to discover the key findings of the work and how SmartChemistry® from deepmatter® was able to add value to the research.  

The research 

The work carried out by Hirst’s team centred around the optimisation of Buchwald-Hartwig coupling animations with different solvents. The data from these reactions was recorded using the DeviceX™ probe and collected on the cloud-based DigitalGlassware platform, part of the SmartChemistry® laboratory solution from deepmatter®. Twelve sensors were used in total to monitor different variables, including temperature, colour and pressure, with colour proving to be a particularly beneficial predictor of product formation for this reaction.  

Using sensors inside organic chemistry reactions helps to generate an abundance of data which can be used to develop machine learning tools and augment the process of predicting synthetic routes. These sensors and the data they gather can be seen as a step towards automating, quantifying and more accurately recording findings without the need for a chemist to manually analyse these elements, offering a more streamlined approach to experimentation.  

With SmartChemistry®, the data was stored in an easily accessible cloud-based platform that enabled the algorithm to manage the data sets while still offering transparency and analytical insight to the researchers. 

Three of the most common methodologies used to capture reaction progress are HPLC, NMR and chromatography. While these are extensively used to monitor outcomes in chemical reactions, they use expensive instrumentation which requires continuous upkeep, and the data collected often requires expert interpretation. Additionally, the data is not centrally located, so chemists must manually collate the figures to provide an accurate analysis. 

However, when enough data is collected from a reaction, AI technology can take this data and monitor the product formation, using this information to make subsequent predictions. Plus, the technology can inform the chemist when a reaction is nearing completion or whether it has strayed from expectation, eliminating the need for a chemist to be constantly monitoring the reaction themselves.  

The result

The results of the experiment found that the AI model accurately predicted yield and purity at runtime for the reaction on the whole, reliably fitting the predictive curve. However, large changes in concentration or temperature could affect the rate of reaction in unpredictable ways, so it’s important to understand the way in which more significant changes could affect the accuracy of the predictive model. 

The data collected by SmartChemistry® proves the benefits of the development of machine learning models which can extract and interpret patterns in large-scale data sets, enabling more accurate predictions by creating generalisations. The use of sensors, combined with a cloud-based data analysis solution, offered researchers at the University of Nottingham an exciting opportunity for understanding the potential predictive abilities of AI in chemical experimentation. 

Discover more about the experiment at the University of Nottingham in the study, published in the Journal of Molecular Graphics and Modelling, here. For more information about SmartChemistry®, click here.