Machine learning-based enzyme engineering of PETase for improved efficiency in plastic degradation

Globally, nearly one million plastic bottles are produced every minute. These non-biodegradable plastic products are composed of polyethylene terephthalate (PET). In 2016, researchers discovered PETase, an enzyme from the bacteria Ideonella sakaiensis that breaks down PET and nonbiodegradable plastic. Temperatures above 60 – 65 °C are optimal for PET degradation as the polymer chain fluctuates in this range, allowing water molecules to enter and weaken the chains. However, PETase has low efficiency at these temperatures, thus limiting its usage. Here, we optimized the rate of PET degradation by PETase by designing new mutant enzymes that could break down PET much faster than PETase. We used machine learning-guided directed evolution to modify PETase 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 machine learning-guided directed evolution. Our algorithm generated hundreds of mutants of PETase and screened them using Random Forest to select mutants with the highest Topt. After 1000 iterations, we produced a new mutant of PETase with Topt of 71.38 °C. We also produced a new mutant enzyme after 29 iterations with Topt of 61.3 °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. Using this approach and novel algorithm, scientists can optimize additional enzymes for improved efficiency.