Upcycling your waste plastics: discovering value-added molecules and interpreting their upcycling pathways
Written by: Chin-Fei Chang, Chemical Engineering, Lehigh University
Recycling waste plastic has been a trending topic for decades. Consumers can easily find products made of recycled plastic in a wide range of merchandise, such as clothes, furniture and building materials. Nowadays, instead of recycling, industries and researchers are seeking to ‘upcycle’ the waste plastic. Upcycling aims to create more added values to the manufacturing process and/or final product. In lieu of giving the waste plastic into its original polymer and producing products made of the same polymer, upcycling refers to decomposing the original polymer into other chemical compounds that are feedstock materials for other high-value products.
Take polyethylene (PE) as an example;, a study1 done by Zhang et. al. showed that PE can be upcycled into long-chain alkylaromatics (a type of chemical compounds that contain one or multiple benzene rings attached by one or more saturated hydrocarbon chains) which traditionally involves an energy-intensive manufacturing process such as naphtha reforming. The proposed tandem reaction routes from PE to aromatics opened up a new possibility of manufacturing aromatics with a lower impact to the environment and higher economic values compared to merely recycling beverages bottles into garbage bags.
This led us to wonder: “Are there any other high-value chemical compounds that the waste plastics can be transformed to?”, “What reaction routes are required to transform the plastic polymer into these molecules?”
To answer these questions, we firstly employed a reaction network generator called RING2. Simply providing RING a reactant- here we used n-decane (the molecule image as shown in Figure 1, a saturated hydrocarbon with 10 carbons)- and a series of reaction rules , it generates thousands of reaction paths and intermediates.
Unfortunately, the catalytic upcycling of plastics can result in a complex reaction network, and it could be daunting to manually analyze the plausible reaction routes to the interested upcycled molecules. This eventually led us to start building a computational workflow that aims to streamline the process of identifying new value-added molecules and elucidating their reaction pathways from polyolefins (a general representation of waste plastics).
In this work, we proposed an automated workflow, which integrates reaction network generation, machine learning-based thermochemistry calculation, and a cheminformatics tool for molecular conformation exploration, to search for plausible value-added aromatics and their most possible flux-carrying pathways, as shown in Figure 1. With the possible flux-carrying pathways elucidated from reactants to final products (the value-added aromatics), it provides the insights of the catalysts selection and reactions planning for the desired final products.
Figure 1. An automated network-screening workflow.
Firstly, by providing information to RING, that are: a model polyolefin molecule -n-decane – and a series of catalytic reaction rules (1) dehydrogenation of a carbon-carbon single bond (the single bond between two carbons becomes a double bond given that one carbon-hydrogen bond of each carbon is broken and the hydrogen atoms leave as hydrogen molecule) and (2) hydrogenation of a carbon-carbon double bond (an inversion of dehydrogenation) that are likely to occur on metal sites, (3) β-scission (a cracking process of hydrocarbons) likely to occur on an acid site, (4) isomerization (molecular fragment transform into an “isomer” which has a different chemical structure, but the number of atom remains the same) leading to methyl branching, plausibly on an acid site, and (5) cyclization (the linear part of a hydrocarbon molecule that contains at least 6 carbons transform into a cyclic hydrocarbon molecule) also occurring on the acid sites of ɣ-Alumina, we generated a reaction network consisted of over 3,700 species and 24,500 reactions. Within the generated, giant reaction network, we applied a query function built in RING to identify 78 aromatic molecules and the dehydroaromatization reaction pathways from n-decane toward each molecule. We recognized that all aromatic molecules found involve a sequence comprising dehydrogenation, β-scission, and cyclization steps (in slightly different order).
To evaluate the reaction pathways found for the 78 aromatics, we calculated the thermochemical properties of the species involved using a cheminformatic tool, RDKit3, and a neural network-based energy calculator, TorchANI4. Through the machine learning based estimation on molecular potential energy, it was possible to process massive amounts of molecules while the accuracy is preserved. The computational cost for the molecular properties estimation is about 2.4 minutes per molecule.
Once the thermochemical properties of molecules found in the pathways (pathways that start from n-decane to aromatics) are computed, we can analyze the Gibbs free energy change of each pathway using the change of reaction enthalpy and entropy at T(temperature) = 298K. The Gibbs free energy (ΔG) can be thought of as an indicator whether a reaction will occur spontaneously or not . For reactions that are nonspontaneous, it usually requires an external input (or energy) for the reaction to occur. Each reaction step in a reaction pathway is denoted with its Gibbs free energy change (unit: kJ/mol) as the values shown next to the arrows in Figure 2. Knowing the thermochemical properties of each pathway, we can identify the plausible flux-carrying pathway for each aromatic compound using a ‘min-max’ technique. For example, there are 461 reaction pathways that a reactant n-decane can go through to transform into benzene (considering that the length of pathway cannot exceed 2 more steps than the minimal length of pathway). Within the 461 reaction pathways, we firstly marked down the reaction step with the highest ΔG, i.e. ΔG_max for each pathway. Next, the pathways with the lowest ΔG_max are chosen and called the ‘min-max’ pathways. The step with this lowest energy value is identified as the bottleneck reaction (shown in a dotted square in Figure 2). This process of finding ‘min-max’ pathways is repeated for the remaining steps in each of these pathways until one final pathway is chosen. This final pathway is referred to as the ”plausible flux-carrying pathway”. The most plausible flux-carrying pathway for 7 common aromatic feedstocks are illustrated in Figure 2, where you might notice that different aromatic products can be derived from the same intermediates.
Figure 2. The best pathway from n-decane to benzene, toluene, xylenes, styrene, and ethylbenzene. All best pathways shared the same bottleneck reaction; different aromatic products can be reached through branched pathway from certain intermediates (3-decene or 3,5-decadiene for example)
In summary, the workflow is system-agnostic and can be used to understand complex reaction networks, such as those in catalytic upcycling systems. We believe that this workflow can also be applied to uncover pathway-level information in various other upcycling chemistries. Additionally, it highlights the importance and economic value of upcycling, drawing broader attention to its benefits.
This article aims to provide a general overview of the published work for the general audience. Details of the computational workflow, methods and results can be found in the full article5.
Reference:
[1] Zhang, F.; Zeng, M.; Yappert, R. D.; Sun, J.; Lee, Y.-H.; LaPointe, A. M.; Peters, B.; Abu-Omar, M. M.; Scott, S. L. Polyethylene Upcycling to Long-Chain Alkylaromatics by Tandem Hydrogenolysis/Aromatization. Science 2020, 370, 437–441.
[2] Rangarajan, S.; Bhan, A.; Daoutidis, P. Language-Oriented Rule-Based Reaction Network Generation and Analysis: Description of RING. Comput. Chem. Eng. 2012, 45, 114–123
[3] Landrum, G.; Tosco, P.; Kelley, B.; Ric,; sriniker,; gedeck,; Vianello, R.; Schneider, N.; Kawashima, E.; Dalke, A. et al. Rdkit/Rdkit: 2021 09 4 (Q3 2021) Release. 2022; https://doi.org/10.5281/zenodo.5835217, (accessed Jan 27, 2022).
[4] Gao, X.; Ramezanghorbani, F.; Isayev, O.; Smith, J. S.; Roitberg, A. E. TorchANI: A Free and Open Source PyTorch-Based Deep Learning Implementation of the ANI Neural Network Potentials. J. Chem. Inf. Model. 2020, 60, 3408–3415.
[5] Chang, C.-F.; Rangarajan, S. Machine Learning and Informatics Based Elucidation of Reaction Pathways for Upcycling Model Polyolefin to Aromatics. Journal of Physical Chemistry A 2023, 127 (13), 2958–2966.
