Executive Summary : | since the discovery of polyethylene terephthalate (PET) degrading enzyme PETase, various hydrolases have been reported to degrade polyester plastics into monomers. This process is a green and promising alternative to plastic waste pollution and contributes to the economy. However, industrial scale processes face challenges such as low thermostability of enzymes and slow rate of ester bond cleavage. Recently, a successful PETase variant called FAsT PETase was engineered using Machine Learning (ML) techniques, showing the potency of the method. Approximately 300 enzymes in the hydrolase superfamily show differential reactivity, suggesting that anchor residues on the surface of the active site cleft play a major role in plastic degradation. It is nearly impossible to synthesize and investigate all protein sequences from the hydrolase superfamily on all plastic variants. To overcome this challenge, deep learning models can predict the binding free energy (BFE) of hydrolase enzyme-plastic pairs. All-atom MD simulations on extensive conformational searches can be used for free energy calculations, but due to high computational demands, BFE calculations for the entire hydrolase superfamily-plastic pairs are practically unviable. ML approaches can predict BFEs of unknown enzyme-plastic pairs by learning the correlation between computed BFE from all-atom MDs and descriptors given by structural metrics. Incorporating state-of-the-art methods for calculating binding free energy as an input feature may improve the accuracy of BFE predictions on unknown hydrolase-plastic pairs. |