GPS-MBA: computational analysis of MHC class II epitopes in type 1 diabetes

Abstract

As a severe chronic metabolic disease and autoimmune disorder, type 1 diabetes (T1D) affects millions of people world-wide. Recent advances in antigen-based immunotherapy have provided a great opportunity for further treating T1D with a high degree of selectivity. It is reported that MHC class II I-A(g7) in the non-obese diabetic (NOD) mouse and human HLA-DQ8 are strongly linked to susceptibility to T1D. Thus, the identification of new I-A(g7) and HLA-DQ8 epitopes would be of great help to further experimental and biomedical manipulation efforts. In this study, a novel GPS-MBA (MHC Binding Analyzer) software package was developed for the prediction of I-A(g7) and HLA-DQ8 epitopes. Using experimentally identified epitopes as the training data sets, a previously developed GPS (Group-based Prediction System) algorithm was adopted and improved. By extensive evaluation and comparison, the GPS-MBA performance was found to be much better than other tools of this type. With this powerful tool, we predicted a number of potentially new I-A(g7) and HLA-DQ8 epitopes. Furthermore, we designed a T1D epitope database (TEDB) for all of the experimentally identified and predicted T1D-associated epitopes. Taken together, this computational prediction result and analysis provides a starting point for further experimental considerations, and GPS-MBA is demonstrated to be a useful tool for generating starting information for experimentalists. The GPS-MBA is freely accessible for academic researchers at: http://mba.biocuckoo.org.

Figure 1. A schematic diagram for the adapted Gibbs sampling approach which was used to determine the nonamer core peptides for I-A^g7 and HLA-DQ8 epitopes.

Figure 2. Screen snapshot of the GPS-MBA 1.0 software.

The default threshold was chosen for MHC Class II I-A^g7 (Medium). As an example, the prediction results for the Mouse Igκ chain C region (UniProt ID: P01837) are shown.

Figure 3.The. performance evaluation of GPS-MBA 1.0 and a comparison with other approaches.

The LOO validation and 4-, 6-, 8- and 10-fold cross-validations were performed for (A) mouse I-A^g7 and (B) human HLA-DQ8, respectively. We compared the performance of GPS-MBA 1.0 with MHC2Pred and RANKPEP using the LOO validation for (C) I-A^g7 and (D) HLA-DQ8, respectively. We also performed a cross-evaluation by using the HLA-DQ8 predictor in GPS-MBA to predict (C) I-A^g7 epitopes and (D) vice versa.

Figure 4. The search options for the TEDB 1.0 database.

(A) Users are able to simply input ‘RAN’ and select “Gene Name” for querying. (B) The results are shown in a tabular format. Users can then click on the TEDB ID (TEDB-HS-00015) to visualize the detailed information. (C) The detailed information on human RAN. The experimentally identified and predicted epitopes are presented.

Figure 5. The sequence logos of the core nonamers for (A) I-A^g7 and (B) HLA-DQ8.