Integrating COSMO-Based σ-Profiles with Molecular and Thermodynamic Attributes to Predict the Life Cycle Environmental Impact of Chemicals

Abstract

Life Cycle Assessment (LCA) has become the main approach for the environmental impact assessment of chemicals. Unfortunately, LCA studies often require large amounts of data, time, and resources. To circumvent this limitation, here we propose a streamlined LCA method that predicts the impact of chemicals from molecular descriptors, thermodynamic properties, and surface charge density distributions of molecules (COSMO-based σprofiles). Our approach uses mixed-integer nonlinear models to automatically construct predictive equations of the life cycle impact of chemicals from a set of attributes that are more accesible than full LCA inventories. We applied our method to predict the life cycle impact of 90 chemicals from three attribute sets: 15 molecular descriptors, 12 thermodynamic properties, and discretized σ-profiles. Nine impact categories were estimated, including among others the Global Warming Potential and Eco-Indicator99. Results show that models based on molecular and σ-profile attributes show similar performance to those based on molecular and thermodynamic attributes. This facilitates the application of streamlined LCA when developing new chemicals and processes, avoiding the experimental determination of thermodynamic properties. Furthermore, molecular, thermodynamic, and σ-profile attributes used together provide the most accurate predictions. Overall, this work aims to enhance chemical environmental assessment, facilitating their screening and enhancing the development of more sustainable processes and products.

Group contribution‐based LCA models to enable screening for environmentally benign novel chemicals in CAMD applications

“…Nevertheless, the models proposed here have comparable performance to the respective state-of-the-art models. The CoD of the GWPjGC lies within the range of the linear and nonlinear models (from negative values to 0.37) developed by Wernet et al, 10 the nonlinear models of the FineChem tool (0.18-0.64) developed by Wernet et al, 9 and the nonlinear models (0.25-0.30) developed by Kleinekorte et al 14 Similar values are also obtained when these models are compared in terms of μARE (i.e., 62% in this work compared to 58% in the work of Wernet et al, 9 50%-88% in the work of Song et al, 11 and 30%-50% in the works of Calvo-Serrano et al 12,13 ). It should be noted that with respect to correlation between model estimations and target values for the testing phase, some of the previous works report R 2 Pearson and Spearman metrics instead of CoD.…”

“…The CoD of the CEDjGC lies within the range of the linear models (0.01-0.43) developed by Wernet et al 10 but is smaller than the values reported by the FineChem tool (0.45-0.71) developed by Wernet et al 9 In terms of μARE, the CEDjGC performance is somewhat worse when compared to the 22%-30% error reported in the works of Wernet et al 9,10 and the 16%-31% error reported in the works of Calvo-Serrano et al 12,13 but it has similar performance when compared to the model proposed by Song et al 11 In terms of R 2 Pearson and Spearman metrics, the CEDjGC model in this work achieves an R 2 Pearson of 0.55, which is somewhat better than the range (0.45-0.52) reported by Song et al, 11 while the R 2 Spearman in the works of Calvo-Serrano et al 12,13 ranges from 0.5 to 0.6.…”

“…[6][7][8] Two pioneering works in this direction are those of Wernet et al 9,10 who proposed a method to estimate the life cycle impact of chemicals from molecular descriptors, using neural networks and principal component analysis (PCA) for reducing the dimensionality of the predictor variables. Following the concept of this approach, some recent studies proposed improvements by selecting different molecular descriptors and neural network structures, 11 adding thermodynamic properties and selecting predictor variables based on mixed integer programming, 12 integrating COSMO-based σ-profiles, 13 adding process descriptors, 14 and introducing classification methods and new data for novel biorefinery concepts. 15 Kleinekorte et al 16 summarized the performance of these relevant methods to estimate life cycle impacts on the basis of molecular, property and process descriptors, ahead of collection of life cycle inventories.…”

Life Cycle Assessment for the Design of Chemical Processes, Products, and Supply Chains

“…The logic of these models is that the molecular structure directly influences the complexity of its production process and its hazard and fate at the end of life, and thereby its life cycle environmental impacts. This correlation between molecular structure and the LCIA can be fitted by multi-linear regression models (54)(55)(56) or nonlinear models, such as artificial neural networks (57)(58)(59)(60). Herein, the physical properties of the product, e.g., the molar mass or the number of functional groups, are used as input to describe the chemical of interest.…”

Process modelling and life cycle assessment coupled with experimental work to shape the future sustainable production of chemicals and fuels

“…Bearing these limitations in mind, some authors developed streamlined LCA methods to predict the impact embodied in chemicals from information readily available in practice, thereby avoiding the need for process models. [146][147][148] These approaches work well for specific chemicals, mainly petrochemicals, while their application to a broader set of molecules is yet to be studied. Another topic that remains largely unexplored is the multi-objective molecular design of chemicals with minimum environmental footprint.…”