Machine Learning-Based Enzyme Engineering of PETase for Improved Efficiency in Degrading Non-Biodegradable Plastic

Abstract

Globally, nearly a million plastic bottles are produced every minute (1). These non-biodegradable plastic products are composed of Polyethylene terephthalate (PET). In 2016, researchers discovered PETase, an enzyme from the bacteria Ideonella sakaiensis which breaks down PET and nonbiodegradable plastic. However, PETase has low efficiency at high temperatures. In this project, we optimized the rate of PET degradation by PETase by designing new mutant enzymes which could break down PET much faster than PETase, which is currently the gold standard. We used machine learning (ML) guided directed evolution to modify the PETase enzyme to have a higher optimal temperature (Topt), which would allow the enzyme to degrade PET more efficiently. First, we trained three machine learning models to predict Topt with high performance, including Logistic Regression, Linear Regression and Random Forest. We then used Random Forest to perform ML-guided directed evolution. Our algorithm generated hundreds of mutants of PETase and screened them using Random Forest to select mutants with the highest Topt, and then used the top mutants as the enzyme being mutated. After 1000 iterations, we produced a new mutant of PETase with Topt of 71.38{degrees}C. We also produced a new mutant enzyme after 29 iterations with Topt of 61.3{degrees}C. To ensure these mutant enzymes would remain stable, we predicted their melting temperatures using an external predictor and found the 29-iteration mutant had improved thermostability over PETase.Our research is significant because using our approach and algorithm, scientists can optimize additional enzymes for improved efficiency.