Predicting protein-protein interactions in E. coli using machine learning methods

The multitude of protein-protein interactions allows an organism to function and maintain cellular homeostasis. Several high-throughput techniques are employed to decipher complete interaction network of a cell and to understand its biology. However, experimental techniques are flawed by the large amount of noise and are limited by the lack of coverage. Computational techniques are therefore sought to predict genome-wide protein-protein interactions. In silico approaches mainly use one or the other genome-context methods to identify the interacting protein pairs. Machine learning algorithms trained on physicochemical characteristics of the proteins have also been used to determine interacting partners. In this work, we have used combined genome context methods as data features to train machine learning algorithms. We observed through several numerical experiments that NN performs better than SVMs on a known dataset. We also aim to predict genome wide protein interaction network for different organism using the best model and efficient algorithm.