PPIBA-D: Protein Protein Interactions Binding Affinity Through Deep

Abstract

Protein-Protein interactions(PPI) are vital effective working for the machinery in organisms and are important across biological activities. In this work, deep learning and machine learning models were used for predicting binding affinities of PPI. Both machine learning and deep learning algorithms were implemented in python. Six classes namely antigen-antibodies, enzyme-inhibitors, G-proteins, receptors, other-enzymes and mis- cellaneous were used. Autocovariance and Autocorrelation for all six classes are calculated followed by predicting the binding affinity and dissociation constant. 1048 protein complexes con- taining various protein sequences made up the dataset utilized for training, validating and testing the Logistic Regression , Gaussian Naive Bayes , GridSearchCV, ADABoost, Random Forest, Decision Tree, SVM, KNN, Deep Belief Network, CNN, RNN, LSTM. The overall accuracy of the system was about 95.59 percentage. The LSTM technique has been used to successfully identify unique binding affinity on a web server known as PPIBA- D.