The Predication Of Immunoproteasome Cleavage Sites

Since the proteolysis of protein by the immunoproteasome is the first step of MHC-Ipathway, this reaction actually gives the MHC-I epitopes some basic characteristics.Therefore, integrating the result of proteasome cleavage prediction into the prediction ofMHC-I epitopes will improve the accuracy of the latter.Based on the biological process, our research analyzes the cleavage preference ofactive sites in immunoproteasome and crucial physicochemical properties of amino acidside-chain involved in the reaction. Due to lack of necessary biological backgroundknowledge, we use non-linear machine learning method to mine the immunoproteasomecleavage data. Different from the encoding in the former method in the field, we use thekey physicochemical properties of amino acid side-chain described above to encode thedata. Our way significantly reduces the dimensions of feature space, while maintains highaccuracy of prediction.Compared with the mainstream tool--netchop3.0, our model owns a significantlybetter performance no matter on training set or testing set. Specifically, our modelachieves a sensitivity of97.7%and a specificity of90.6%on training set.Our research innovatively establishes three predication models subject to three kindsof cleavage active sites in immunoproteasome. Through comparison of the threesubmodels and the omnipotent model, we found although the multimodels can improvethe predication performance, the improvement is much smaller than our expectation.There are two possible reasons. First, the variances of the sequence patterns of thesubstrate constraint by the enzyme active sites are not significant. Second, the criterion todivide the submodels may still need improved.In addition, by the correlation analysis of the crucial cleavage site P1with otherpositions, we acquire potential anchor positions on substrate sequence and the quantityrelationship between the physicochemical properties, which may contribute to furtherresearch. We discovery there are some potential anchor positions between the true andfalse samples, however, the false samples usually lack some anchor positions. It indicates the missing of the key amino acids on the supposed anchor positions may result in themissing of cleavage.