Machine learning-based structural classification of lytic polysaccharide monooxygenases

Lytic polysaccharide monooxygenase (LPMO) is a copper-dependent redox enzyme and according to CAZy is classified either as cellulolytic or chitinolytic. According to CAZy, there are eight families of LPMO namely AA9, AA10, AA11, AA13, AA14, AA15, AA16, and AA17, where AA stands for Auxiliary Activity. Previously, using the sequence information machine learning-based functional annotation was successfully completed using neural network and LSTM models. This was done for AA9 and AA10 as the number of sequences was large enough to train a model with high-performance indices. Here, the goal is to use existing 3D structures and AI-based models of the remaining LPMO sequences and train a machine learning model to identify and classify LPMO with the help of structural features by performing either a neural network algorithm or other suitable methods. Using the features of LPMO such as surface depth, accessible area, electrostatic charge distribution, and geometric features (independent features that define the shape and are not based on enzyme reaction mechanism) we will identify the features with high signal-to-noise ratio/significance using ensemble feature selection method. The features were extracted using Pymol, MdTraj, Biopython, Open3d, and Bio3D tools from OBJ and STL format. The model thus trained using structural features will enable identifying and annotating newer LPMO sequences belonging to one of the eight AA families.