To generate high quality training data for our MHC binding and presentation prediction models, we stably transfected alleles into HLA-null, K562 to create mono-allelic cell lines. Then, we performed immunoprecipitation using W6/32 antibody and gently eluted the peptides. Finally, they were analyzed using LC-MSMS. To validate the performance of our prediction algorithms, we also processed 12 lung and colorectal tumor tissues with the same protocol.