Machine Learning Methods for Predicting 3D Structure of Proteins

The prediction of the three-dimensional structure of the protein is one of the most discussed problems in bioinformatics at the moment. Knowledge of the structure is necessary for understanding the functionality of the protein, which in turn can help in the fight against diseases against which there are no drugs. The aim of this work is to predict the protein contact map. Using the information on the limitations contained in the contact map, it is possible, under certain circumstances, to reconstruct the three-dimensional structure of the protein with high accuracy, due to the fact that this matrix is a two-dimensional representation of the protein. In the course of work, various machine learning models were considered (XGBoost, CatBoost, Logistic Regression, CNN, ResNet, BiLSTM, LSTM). As a result, it was possible to achieve a value of 0.79 on the F1 metric using a hybrid consisting of ResNet and BiLSTM.