AI3SD Project: Active learning for cost-efficient reaction prediction using kinetic data

Abstract

Predicting an optimal set of conditions for a given reaction is a challenging task. This becomes even more challenging when also trying to predict the performance of a reaction, such as the final yield of product or how long it will take. Existing approaches to this problem use machine learning algorithms that are fed a large volume of single timepoint yield data (the amount of product generated after a set time). Using this data is problematic; if a reaction only reached 50% would it reach 99% if run for longer? How much longer? Was there a slow catalysts activation period followed by a rapid reaction? Or was a catalyst poisoned, stopping the reaction midway? These are factors described by the reaction kinetics. We propose that using reaction kinetic data will lead to better predictive models. However, collecting kinetic data is significantly more costly and time consuming than collecting single timepoint yields. To minimise these issues, we will be using a cutting-edge machine learning technique called active learning. Rather than performing every possible reaction combination to build our model, active learning first picks the most important set of conditions. The user collects this data, and the
active learning process is updated and begun again, resulting in a highly cost-efficient procedure minimising experimental requirements.

More information

Identifiers

ORCID for Samantha Kanza: orcid.org/0000-0002-4831-9489

ORCID for Jeremy G. Frey: orcid.org/0000-0003-0842-4302

ORCID for Mahesan Niranjan: orcid.org/0000-0001-7021-140X

Catalogue record

Export record

Altmetrics