Utilizing modular neural networks to predict MHC class II-binding peptides

In order to minimize the number of peptides required to be synthesized and to advance the understanding for the immune response, some researchers have applied the models based on traditional artificial neural networks to predict which peptides can bind to a specific MHC molecule. However, there is still some space for the improvements of the models in learning speed and generalization ability. It has been observed that modular neural networks outperform the single neural networks in diverse domains. Thereupon, the modular neural networks are introduced to predict MHC II-binding peptides for the first time. Compared with the models based on single artificial neural networks, modular neural networks are empirically proved to be more effective and the test results show an obvious increase in the accuracy rates of prediction (approximately 11-37%) for the peptides to bind or not bind to HLA-DR4 (Bl *0401).