Abstract

Towards Big Data in Flow Chemistry: Popularity Trends and Best Practices

Maarten Dobbelaere, PhD Candidate, Ghent University

Machine learning can speed up organic chemistry tasks such as outcome prediction and condition recommendation [1]. However, current reaction databases suffer from different problems that hinder the flow chemist from using machine learning efficiently. First and foremost, the major part of chemical literature is not (easily) accessible [2]. Next, the accessible data is often incomplete and biased towards specific reactions and conditions [3]. Most importantly, all databases contain nearly only data from batch processes [4]. Since conditions in continuous-flow processes are substantially different from batch, the current data resources can hardly be applied by flow chemists. In this work, chemical reaction data is extracted from all flow chemistry papers available on Web of Science, until 2022. The trends in the flow chemistry database are compared with data available in current reaction databases, in terms of reactants, product types, reaction classes, and reagents. The presentation has following goals: (i) explain the main discrepancies between the available databases and flow data, (ii) visualize the popularity trends of reaction types and conditions in flow and batch over the years, (iii) provide best practices to facilitate FAIR data in flow chemistry.