The ever-growing peptide knowledge promotes the improvement of HLA class I peptide-binding prediction.

•We report the feedback effect of peptide data on a matrix-based prediction method.•We implemented the weighted and unweighted models of this prediction method.•We evaluated the relative performance of the two models on benchmark datasets.•Improvements on both models were obtained by optimizing the training dataset.•The increased number of binders required the modification of the predictive engine.