Rule Induction for Prediction of MHC II-Binding Peptides

摘要

Prediction of MHC (Major Histocompatibility Complex) binding peptides is prerequisite for understanding the specificity of T-cell mediated immunity. Most prediction methods hardly acquire understandable knowledge. However, comprehensibility is one of the important requirements of reliable prediction systems of MHC binding peptides. Thereupon, SRIA (Sequential Rule Induction Algorithm) based on rough set was proposed to acquire understandable rules. SRIA comprises CARIE (Complete Information-Entropy-based Attribute Reduction algorithm) and ROAVRA (Renovated Orderly Attribute Value Reduction algorithm). In an application example, SRIA, CRIA (Conventional Rule Induction Algorithm) and BPNN (Back Propagation Neural Networks) were applied to predict the peptides that bind to HLA-DR4(B 1*0401). The results show the rules generated with SRIA are better than those with CRIA in prediction performance. Meanwhile, SRIA, which is comparable with BPNN in prediction accuracy, is superior to BPNN in understandability.