Characterisation of essential proteins in proteins interaction networks

Abstract

The identification of essential proteins is theoretically and practically important as it is essential to understand the minimal surviving requirements for cellular lives, and it is fundamental of drug development. As conducting experimental studies to identify essential proteins are both time and resource consuming, here we present a computational approach in predicting them based on network topology properties from protein-protein interaction networks of Saccharomyces cerevisiae, Escherichia coli and Drosophila melanogaster. The proposed method, namely EP3NN (Essential Proteins Prediction using Probabilistic Neural Network), employed a machine learning algorithm called Probabilistic Neural Network as a classifier to identify essential proteins of the organism of interest. EP3NN uses degree centrality, closeness centrality, local assortativity and local clustering coefficient of each protein in the network for such predictions. Results show that EP3NN managed to successfully predict essential proteins with an average accuracy of 95% for our studied organisms. Results also show that most of the essential proteins are close to other proteins, have assortativity behaviour and form clusters/sub-graph in the network.