Pioneering Enzyme Classification by Combining EDA with Sequential Neural Networks

This study presents an innovative method for enzyme classification by integrating Exploratory Data Analysis (EDA) with a Neural Network (NN) framework. Utilizing a dataset of 858,777 annotated amino acid sequences from ten different organisms, the approach predicts enzyme categories in a test set of 253,146 samples, excluding sequences that exceed a predefined length. The preprocessing steps involve removing infrequent amino acids (X, U, B, Z) and addressing amino acids (B, Z) present only in the training set. The Neural Network model, featuring an embedding layer, bidirectional LSTM layers, and a dense output layer, achieves a notable accuracy of 79% on the test data after 20 epochs. The model's performance is further evaluated with a detailed classification report and confusion matrix across 20 enzyme categories. This research highlights the effectiveness of combining EDA with NN techniques in advancing enzyme classification in bioinformatics and molecular biology. Future research will focus on incorporating additional features and optimization techniques to further enhance the model's performance.