We consider the problem of how digital twins can improve the process of synthesising catalysts so that the carbon monoxide conversion efficiency of chemical reactors can be improved in general. Six key steps in the life cycle of such a digital twin are outlined, and new results are presented for the first two steps, viz. conceptualisation and model development, using published data from 96 well-referenced publications on such chemical reactors. A range of machine learning algorithms is considered as a candidate for developing the digital twin that can accurately predict the conversion efficiency of the chemical reactor for a variety of operational conditions. It is shown that Gaussian process regression models outperform linear regression models, support vector machine models, decision trees and ensemble trees in this context.