Improved Machine Learning Models by Data Processing for Predicting Life-Cycle Environmental Impacts of Chemicals

Abstract

Machine learning (ML) provides an efficient manner for rapid prediction of the life-cycle environmental impacts of chemicals, but challenges remain due to low prediction accuracy and poor interpretability of the models. To address these issues, we focused on data processing by using a mutual information-permutation importance (MI-PI) feature selection method to filter out irrelevant molecular descriptors from the input data, which improved the model interpretability by preserving the physicochemical meanings of original molecular descriptors without generation of new variables. We also applied a weighted Euclidean distance method to mine the data most relevant to the predicted targets by quantifying the contribution of each feature, thereby the prediction accuracy was improved. On the basis of above data processing, we developed artificial neural network (ANN) models for predicting the life-cycle environmental impacts of chemicals with R 2 values of 0.81, 0.81, 0.84, 0.75, 0.73, and 0.86 for global warming, human health, metal depletion, freshwater ecotoxicity, particulate matter formation, and terrestrial acidification, respectively. The ML models were interpreted using the Shapley additive explanation method by quantifying the contribution of each input molecular descriptor to environmental impact categories. This work suggests that the combination of feature selection by MI-PI and source data selection based on weighted Euclidean distance has a promising potential to improve the accuracy and interpretability of the models for predicting the life-cycle environmental impacts of chemicals.

Enhanced Deep-Learning Model for Carbon Footprints of Chemicals

“…After data cleaning, they accounted for the PCFs of 547 unique organic chemicals, which are significantly larger than the data sets used in previous studies (Figure S1). ,− , Most data from the industry have not been included in any public LCA databases and have resulted in a structurally more diverse training data set than previous works (Figure S2). In addition, chemicals in our new data set have more widely distributed physicochemical proprieties and PCFs than the chemicals that were previously used for modeling (Figures S1 and S3).…”

“…ML models are usually considered “black boxes” due to their poor interpretability. , In response, previous studies have used a SHAP-based approach to identify the most important molecular descriptors that affect the PCFs of chemicals . However, it could not provide clear insights on, for example, PCF-intensive substructures and raw materials to guide the design of more sustainable molecules and processes.…”

“…If the distance was above the threshold, the query chemical was considered outside the AD; otherwise, it was considered within the AD. The previous pre-LCA tools were mostly developed based on chemicals in ecoinvent or a part of ecoinvent. − Therefore, organic chemicals in ecoinvent used in a previous study were collected to calculate the AD of the previous pre-LCA tools for comparison.…”

“…In 2009, the first molecular structure-based pre-LCA tool, FineChem was developed based on artificial neural networks (ANNs) and basic molecular descriptors to estimate the environmental impact of chemicals. Subsequently, several other pre-LCA tools were proposed based on machine learning (ML) algorithms and mostly data from ecoinvent. − For example, Sun et al developed a new ANN model based on data processing strategies (ANN-DP) to enhance the predictive ability of the model . Song et al trained multilayer ANNs to rapidly estimate the life cycle impacts of chemicals (rapid-ANN) and highlighted the importance of understanding the applicability domain (AD) of models .…”

Networking Chemicals Flows: Efficiency–Value–Environment Functionalized Symbiosis Algorithms and Application

“…Traditional research focuses on the element and energy flows in chemical enterprises or clusters, e.g., Ozalp 15 gave energy and material flow models of hydrogen production via steam reforming of methane in the United States, Chae et al 16 developed a mathematical model to synthesize the waste heat utilization network targeting a petrochemical complex, Tian et al analyzed the carbon metabolism, 17 sulfur metabolism 18 and energy metabolism 19 in a typical fine CIP, Guo et al 20 analyzed the carbon element flows in a natural gas CIP, and Ma et al 5 presented a general chlorine metabolic model at the industrial park level. Since the chemical flows involving multiple units and phases often form a complex network, programming algorithms are effective in related studies, including a genetic algorithm by Wu and Wang, 21 and improved machine learning models by data processing by Sun et al 22 Some studies focused on the network flow analysis algorithms to address the complexity of chemical industrial chains, e.g., Ren et al 23 proposed an algorithm methodology to calculate the resource productivity of crude oil from a complex network perspective.…”