Prediction of MHC class II-binding peptides based on sequential learning

Predicting which peptides can bind to a specific major histocompatibility complex (MHC) molecule could have great value for minimizing the number of peptides required to be synthesized and assayed. Artificial neural networks (ANNs)-based prediction method, which usually adopts backpropagation neural networks (BPNN) as a prediction model has high prediction accuracy, while its learning efficiency is low and its incremental learning cannot be realized. Sequential learning (SL) adds output neurons in such a way that a correct mapping between input and output patterns is guaranteed. We propose the SL-based prediction method, which chooses the modified sequential learning ahead masking (SLAM) model combined with incremental learning (FIL-SLAM) to predict MHC II-binding peptides. For the experimental data composed of 650 peptides to bind or not bind to HLA-DR4 (B1*0401), compared with BPNN, the proposed method shows a significant reduction in consuming time (95%) with only a slight reduction (1.3%) in average prediction accuracy.