Domain-adapted discovery of PET depolymerisation routes using in-house curated scientific literature

Area of Science:

• Chemical Engineering
• Materials Science
• Environmental Science

Background:

• Plastic waste, particularly polyethylene terephthalate (PET), poses a significant environmental challenge.
• Current depolymerisation methods require optimization for industrial scalability and economic viability.

Purpose of the Study:

• To develop a data-guided approach for discovering and evaluating experimentally feasible PET depolymerisation pathways.
• To leverage curated chemical literature and advanced language models for this discovery process.

Main Methods:

• A domain-adapted language model was trained on curated chemical literature.
• The model was used to generate and assess potential PET depolymerisation routes.
• Experimental feasibility of generated pathways was a key evaluation criterion.

Main Results:

• The combined approach successfully generated experimentally feasible PET depolymerisation pathways.
• The data-guided methodology proved effective in identifying viable routes for plastic waste management.

Conclusions:

• This work demonstrates a practical, data-driven strategy for advancing chemical recycling of PET.
• The approach offers a new paradigm for evaluating and discovering efficient plastic waste management solutions.