New feature vectors on prediction of HIV-1 protease enzyme cleavage sites

Protease enzymes which are inside the living organisms, implement many useful biological functions. However, many infectious microorganisms such as viruses and parasites use proteases to be infected as virulence factors. The main task of proteases is to cleave the polyproteins synthesized newly at the appropriate places to make them structural components. In this way, virulent proteases take role in maturation and replication of microorganisms. Hence, unravelling the specificities of proteases is of great importance to develop drugs and vaccines. However, little is known about the cleavage specificities of these proteases. It is therefore, an important challenge to collect experimental protease data and to develop appropriate bioinformatics feature encoding schemes, algorithms to characterize the specificities for all proteases. In this thesis, human immunodeficiency virus type 1 (HIV-1) protease site prediction has been studied.When studying on proteins, there are two basic points considered: physicochemical relationships and the positions of the residues in protein sequnces. These two references are the keys to understand the functions of the proteins and can be used to predict where HIV-1 protease cleave the polyproteins. This hypothesis leads us to develop three feature encoding schemes namely FTKY, BirTVD and BirBOOL to model specificity of HIV-1 protease.For the prediction of HIV-1 protease cleavage sites, peptide samples of two up-to-date HIV-1 protease datasets have been encoded with feature encoding techniques and extracted their features with Principal Components Analysis and Linear Discriminant Anaysis. Subsequently, they have been classified by Linear Support Vector Machines algorithm. According to empirical results obtained, BirTVD and BirBOOL methods have achieved better performance compared to hitherto methods.