The ubiquitin-proteasome system (UPS) plays a critical role in proteolysis and degradation in many physiological processes of the eukaryotes, such as antigen presentation, cell cycle regulation, and transcription factors activation. Recently, due to the importance of selective substrate cleavage of proteasome in the UPS, the cleavage site prediction has attracted considerable interest in computational biology. However, the existing methods are mostly based on nonlinear models with little physicochemical meanings. In this paper, VHSE (Principal component score vector of hydrophobic, steric, and electronic properties), a novel set of amino acid descriptors, was used to characterize the source proteins of 2650 natural MHC class I ligands. Based on the structural descriptions of the amino acids adjacent to the cleavage site, support vector machine (SVM) was then employed to establish the prediction models using linear and RBF kernel functions. A linear SVM model with high prediction capability was obtained, of which the sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and the Matthews correlation coefficient (MCC) were 0.9018, 0.6963, 0.8797 and 0.6131, respectively. The results showed that the hydrophobic, electronic, and steric properties of the amino acids adjacent to the cleavage site are closely related to the selective substrate cleavage, especially for those at the positions of P9, P8, P4, P1, P3', P4', and P5'. The results also showed that hydrophobic potential difference between P1 position and P1'～P5' positions may benefit the cleavage process of the proteasome.

Key words: proteasome, MHC-I ligand, VHSE, SVM, cleavage site