Protein molecules are very important since they take part in many processes in the organisms. Different proteins have different preferences to be involved in various processes, and these preferences determine their functions. The development of efficient and accurate computational methods for determination of the protein functions is of high importance, and therefore this research area is one of the hottest topics in bioinformatics. In this paper we present a method for functionally annotating protein structures. We consider the global characteristics of the protein structure, and also we take into consideration some local characteristics of the binding sites where the inspected protein structure get into interaction with another structure. After extracting the characteristics of the protein structure, then we induce prediction model by using the Binary Relevance method for multi-label learning. We present some experimental results of the evaluation of the method.
Protein function prediction protein structure protein binding site multi-label learning