Prediction of protein interactions by using artificial intelligence techniques based on protein sequence data

Proteins play important roles in all events in the cell and form the basis of biological structures. While carrying out their tasks, proteins communicate with other proteins and molecules in different structures and forms. These structures, called protein-protein interactions, are involved in carrying out all biological processes. Therefore, determining the interactions is an important subject of research. Different methods are used to detect interactions between proteins. These methods are mainly divided into 3 groups as in-vivo, in-vitro and in-silico. Experimental determination of the interactions is evaluated as high throughput studies made in laboratory. These studies require a lot of time and effort. In addition, researchers emphasize that the false positive and false negative rates of such studies are quite high. In order to support these studies conducted in the laboratory environment and to provide them with preliminary information, computational interaction prediction methods are studied extensively. In this study, computational methods have been proposed for the estimation of interactions between proteins. In these methods, feature extraction steps using the protein sequence information from the present databases were generated. The models proposed in this way can be applied to all protein datasets and differ from computational methods that need to know different and complex properties about proteins. By using the obtained feature matrices, interaction prediction was made with a support vector machine based classification system. The performances of the proposed systems were tested with commonly used evaluation criteria. Compared to previous studies, it was seen that the results of the proposed systems on different datasets were successful and acceptable.