(737d) Estimating the Life Cycle Impact of Chemicals from Molecular, Thermodynamic and Charge Density Information Via Mixed-Integer Optimization Techniques

Authors: 
Guillén-Gosálbez, G., Imperial College of Science, Technology and Medicine
Calvo-Serrano, R., Imperial College London
The chemical industry is at present striving to decrease its environmental footprint in the transition towards a more sustainable manufacturing sector. In this context, systematic methods are required to assist in the environmental assessment and optimisation of chemicals, identification of critical hotspots across products’ supply chains and definition of guidelines to effectively retrofit processes so that they adhere to sustainability principles.

The amount of new chemicals produced annually has exponentially increased in the last decades. The database CAS registry1 from Chemical Abstract Service, a division of the American Chemical Society devoted to authoritatively collect disclosed chemical substance information, presently contains more than 120 million organic and inorganic substances and 66 million sequences, with approximately 15,000 new substances being added each day. Hence, an enormous amount of chemical species exists, but yet there is a lack of information regarding their chemical hazards and potential environmental impact. This information, however, is critical for the proper evaluation of their wider sustainability impacts and future market feasibility.

Under the above scenario, the recent trend towards the development of more sustainable products has led to a plethora of environmental assessment tools. From these, Life Cycle Assessment (LCA) has gained popularity in recent years, currently being one of the most extended sustainability assessment methods. LCA is based on the evaluation, for all stages of the life cycle of a product, of all possible interactions of the activities carried out and the environment.

Unfortunately, LCA requires large amounts of data across the products’ supply chain, some of which might be hard to gather in practice. This constitutes one of its main obstacles, and more so in the chemical industry, which contains many complex networks that exchange mass and energy. In these cases, when a full LCA cannot be applied, a simplified version is used instead. These Streamlined LCA (SLCA) approaches follow the same basis as LCA, but generally either simplify the scope of the analysis and/or reduce the amount of information required in the assessment. The precise simplifications to be done (and the assessment discrepancy with the full LCA) have to be specifically considered for the process or activity assessed.

Under these principles, we present herein a novel SLCA approach for the estimation of the impact of chemicals that make use of a set of specific molecular properties. Previous studies demonstrated the prediction capabilities of molecular and thermodynamic attributes2. Here we extend these works by considering the σ-profile of chemicals as attributes. The sigma profile provides the probability distribution of a molecular surface segment having a specific charge density. In this work, we discretize this profile and use it together with molecular and thermodynamic descriptors to estimate eight well known LCA indicators, including the cumulative energy demand (CED), Eco-indicator 99 and global warming potential, among others. The regression models are built using mixed-integer optimization techniques that automatically identify the attributes that best describe the specific impact categories.

We applied our approach to a database consisting of 83 chemicals and considering 15 molecular descriptors, 17 thermodynamic attributes and 8 σ-profile-based descriptors, generating estimates of enough accuracy for the purpose of a standard LCA (i.e. 15-25%). Our framework ultimately leads to simple linear models that can be implemented in computer aided molecular design software, thereby enhancing the application of sustainability principles in the early stages of the development of new chemicals.

References

(1) http://www.cas.org/. Chemical Abstracts Service Home Page http://www.cas.org/ (accessed Oct 18, 2016).

(2) Wernet, G.; Hellweg, S.; Fischer, U.; Papadokonstantakis, S.; Hungerbühler, K. Molecular-Structure-Based Models of Chemical Inventories using Neural Networks. Environ. Sci. Technol. 2008, 42 (17), 6717–6722.