The immune system is initiated and regulated through a process starting from the binding of antigenic peptides to major histocompatibility complex (MHC) molecules. Detailed understanding of such interactions would lead to the development of vaccine design for infectious diseases, and immunotherapies for autoimmune diseases and cancer. Since MHC class 11 genes are highly polymorphic, a computational prediction tool for the first screening of antigenic peptides that bind to MHC class 11 molecules, is highly desirable. In the present study, hidden Markov model (HMM) was applied for the screening of peptides that interact with nine MHC class 11 molecules, specifically, human leukocyte antigen (HLA)-DR1, -DR2, -DR4, -DR7, -DR11, -DR15, -DR17, -DR51, and -DQ2. When high-binding peptides interacting with each MHC molecule were subjected to the constructed HMM model, significantly high likelihood values were obtained, as compared to the non-binding peptides as negative control. With the receiver-operating characteristic analysis for the prediction evaluation, our model showed high prediction accuracy, with an average AUC value of 0.87 for all molecules. The HMM model that was trained by HLA-DQ2 showed significantly low likelihood values to peptides that bind to eight HLA-DR molecules. This suggests the high potency of our HMM model for discriminating HLA-DQ binding peptides from HLA-DR binding peptides. (C) 2003 Elsevier Inc. All rights reserved.