Computational Strategies for Improving PET-Degrading Enzymes: From Atomistic Simulations to AI Stability Prediction
[eng] Plastic pollution remains a critical environmental challenge, with polyethylene terephthalate (PET) being one of the most widely used and persistent plastics. Enzymatic PET recycling offers a promising route toward a circular economy, leveraging biocatalysts to depolymerize PET under mild cond...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universidad de Barcelona |
| Repositorio: | Dipòsit Digital de la UB |
| OAI Identifier: | oai:dnet:ubarcelona__::c3e33e7ded30be9bedfb4d3b5a603185 |
| Acceso en línea: | https://hdl.handle.net/2445/228921 https://hdl.handle.net/10803/697240 |
| Access Level: | acceso embargado |
| Palabra clave: | Dinàmica molecular Enzims Plàstics biodegradables Aprenentatge profund Aprenentatge automàtic Molecular dynamics Enzymes Biodegradable plastics Deep learning (Machine learning) Machine learning |
| Sumario: | [eng] Plastic pollution remains a critical environmental challenge, with polyethylene terephthalate (PET) being one of the most widely used and persistent plastics. Enzymatic PET recycling offers a promising route toward a circular economy, leveraging biocatalysts to depolymerize PET under mild conditions. However, key barriers remain, including enzyme instability at process-relevant temperatures, low depolymerization activity towards post-consumer PET and the resistance of crystalline PET regions to enzymatic attack. This thesis addresses these challenges through an integrated computational strategy combining atomistic molecular simulations and AI-guided protein stability design. First, extensive molecular dynamics simulations were employed to investigate PETase-PET interactions at atomic detail. We developed realistic PET oligomer models representing amorphous and crystalline states, capturing experimentally informed torsional preferences. By applying enhanced sampling techniques such as Hamiltonian Replica Exchange and well-tempered metadynamics, we quantified the energetic barriers associated with binding and catalysis on crystalline PET, revealing a ~50 kJ/mol penalty relative to amorphous PET. Additionally, simulations of different PETase variants highlighted the relationship between thermostability, active site integrity, and catalytic competence, offering atomistic explanations for experimental observations. Second, we developed PANDORA, an AI-based framework for predicting protein stability from sequence alone. Leveraging the Megascale dataset of folding free energies, PANDORA evolved through successive architectures, from one-hot encoded CNNs to graph neural networks and transformer-based embeddings, culminating in a hybrid model that integrates both sequence and structural information. Applied to PETase variants, PANDORA offered accurate ΔG predictions to complement stabilising mutation design. Together, these complementary approaches deliver mechanistic insights into the molecular determinants of PETase function and provide generalizable computational tools for enzyme engineering. By bridging molecular simulations and AI-based predictions, this work contributes to advancing biocatalytic PET recycling toward industrial feasibility, while also offering frameworks applicable to broader challenges in sustainable materials science and protein engineering. |
|---|