Prediction of interactions between SARS-CoV-2 protein and human protein using machine learning methods

Covid-19, which threatens human health worldwide, was declared an epidemic disease by the World Health Organization (WHO) in 2020. Covid-19 disease has had a devastating effect on humanity due to its spread rate and deadly effect. Covid-19 disease is a new type of corona virus caused by the SARS-CoV-2 virus. In this study, we predicted the interactions of SARS-CoV-2 virus proteins with host (human) proteins using machine learning methods. Determination of pathogen-host interactions in studies against infectious diseases is of great importance in taking precautions against the disease. Pathogen proteins interact with host proteins to invade the host. They can interfere with the normal function of host proteins or even compromise a host's immune system by misdirecting and weakening it. For this, it is of great importance to predict the putative protein-protein interaction between the pathogen-host. Identifying protein-protein interactions helps to find out how virus proteins work, how they replicate, and how they cause disease. In our study, we used a protein interaction dataset consisting of 30,046 negative and 20,365 positive data. To predict protein interactions, we encoded amino acids in the protein sequence using the protein feature coding method. Naive Bayes, Bayes Net, k-Nearest Neighborhood, Linear Support Vector Machines, Radial Support Vector Machines and Multilayer Perceptron were used as classification algorithms. Firstly, the data obtained by protein feature coding methods in the literature are given to the classification algorithms. Then, a new method was developed that includes the frequency, location and chemical properties of amino acids. When the classification results of the developed method and the methods in the literature were compared, more balanced results were obtained. The method we developed according to the experimental results we obtained gave 0.513 accuracy, 0.529 sensitivity, 0.471 specificity, 0.514 precision, 0.514 F1-score, 0.027 Kappa values with Naive Bayes and Bayes Networks classification algorithms.