This site contains resources for tool developers interested in predictions of peptide binding to MHC class I molecules. The links on the sidebar lead to a manuscript describing the resource in detail, a dataset of experimental affinities of peptide to MHC molecules and a description of the framework used for the evaluation of prediction methods.
As described in the manuscript, predictions where obtained from public web-servers for all relevant peptide-MHC affinities in the dataset. The correlation between predicted and measured affinities was evaluated using scatter plots, linear regression and ROC analyses. The evaluation of these external tools can be accessed by name of the method or the MHC allele. As described in detail the manuscript, this is not a fair evaluation of the value of each method, primarily because the data available to each method is highly divergent.
A similar evaluation of prediction performance of three prediction methods routinely used by us was carried out using cross-validation on the dataset. In contrast to the comparison of external predictions, this is a fair evaluation of prediction performance of the three methods, as training and testing data was the same for each method. Again, the evaluations of these three internal methods can be accessed by name of the method or the MHC allele.