Sustainability assessment using local lazy learning: The case of post-combustion CO2 capture solvents

The consideration of sustainability is very important for the assessment of life cycle, environmental, health and safety properties of chemicals used in various applications. The screening of wide ranges of molecular structures, prior to the identification of the optimum and most sustainable options, requires the use of efficient and inclusive predictive models. Group contribution (GC) models are popular for the evaluation of numerous molecular options, but they support the calculation of few properties related to sustainability, while their predictive capabilities are often limited by significant data gaps. To address such challenges, we propose the use of a local learning approach as a means of evaluating sustainability properties for a wide range of molecular structures. Supplementing GC methods with data mining ones, such as local lazy learning approaches and exploiting molecular similarities has a potential to improve the predictive capacity of sustainability indices and offers an alternative when GC methods or empirical models are not available for life cycle assessment (LCA) and environmental, health and safety (EHS) hazard assessment indicators. The proposed approach is applied to predict a set of properties (bioaccumulation, persistence and acute aquatic toxicity) of 101 commercial solvents for post-combustion CO2capture