Prediction of host-pathogen protein interactions by computational methods

Knowledge of the pathogen-host protein interactions in the inter species has a vital prospect for a solution strategy to be developed against diagnosis and treatment of infectious diseases. Modeling interactions between proteins has necessitated the development of computational methods in this field, since detection of interactions by experimental methods is both time-consuming and costly. Computational methods are used in decreasing of the detection time and cost; in addition checking of the false detected interactions via experimental methods. Data scarcity, data inadequacy, and negative data sampling are the common problems of computational methods for used in prediction of pathogen-host protein interaction. In this study, the purpose is that prediction accuracy of the pathogen-host interaction increase and negativeness eliminate because of data inadequacy. Within thisframework, extended network model and location based encoding approaches are proposed. Firstly, the extended network model is created by inspired from the hypothesis of that "integrating the known protein interactions within host and pathogen organisms improve the success of prediction of unknown pathogen-host interactions". Secondly, location based encoding is feature extraction method which is used for encoding of amino acid sequences. One of the important factors is feature which affects success in prediction of pathogen-host interaction within machine learning algorithms. In biological databases, the most data is the information of amino acid sequence regarding proteins. Prediction accuracy of pathogen-host interaction will be increased by that a robust feature extraction method is developed on the basis amino acidsequence. Furthermore, extraction of feature vectors for all the known interactions are provided in easier way by the sake of using the information of amino acid sequence. In this thesis, PROSES (Protein SequencebasedEncodingSystem) which is a user-friendly interface and freely accessible web server, has been designed for researchers, who are working on the field of protein encoding and prediction of protein interaction. The web server is especially useful for those who are not familiar with programming languages. PROSES is currently being used at http://proses.yalova.edu.tr which is storedin the web server of Yalova University.