A framework for machine learning assisted directed evolution of plastic-degrading enzymes

Globally over 381 million tonnes of plastic are produced every year. The UK recycling industry is valued at over £27bn and is expected to grow at 5.8% per annum. Polyethylene terephthalate plastics (PET) are a highly recyclable form of plastics with high levels of collection for processing in certain countries (53% of all bottles and 38% of other packaging in the UK). Despite this, mechanical recycling of plastics is an expensive process and cannot be used indefinitely owing to accumulation of contaminants. S&P Global Platts estimates that recycled plastics cost an extra £57 per tonne than virgin plastic cost to manufacture.

Biomanufacturing presents a new and sustainable alternative to the problem of cost-effective, low-energy and carbon-neutral plastic recycling. Plastic-degrading enzymes have gained increasing attention as a solution to a sustainable plastic recycling process. However, naturally occurring enzymes are not well-suited for industrial plastic degradation applications due to the limitations on thermostability and catalytic activity.

Entropix aims to overcome these limitations. The project will demonstrate how modern biomanufacturing technologies paired with industrial AI can deliver new recycling technologies for the depolymerization of plastic to its original raw materials.

Building on Entropix's existing technology stack and working with the University of Liverpool's Virtual Engineering Centre the project will deliver a new made-for-manufacturing machine learning approach to produce directed evolution industrial enzymes engineered to produce better catalytic efficiency and stability for plastic depolymerization, for use in industry and recycling centres in Liverpool and elsewhere.

This innovative development will apply knowledge from different technology sectors, with the application of AI to the process of directed evolution. Past work has shown that machine learning is especially effective at identifying patterns in DNA sequences. Machine learning tools seem well-placed to contribute to this type of practical problem.

Directed evolution identifies the enzyme variants with desired properties and functionalities using iterations of mutation and selection. It has been successfully deployed but is limited by the enormous number of potential mutations for a given protein.

This project will deploy a novel machine learning (ML) directed evolution framework that uses ML approach to carefully guide the selection of candidate mutants at each evolution loop.

The engineered plastic-degrading enzymes are aimed at better cost efficiency and feasible manufacturing. It will integrate with the plastic treatment and recycling industrial market and support the development of a sustainable plastic value chain in the Liverpool region and beyond.