Introduction to DeepB³P

DeepB³P is a transformer-based deep learning model for the prediction blood-brainpenetrating peptides. In DeepB³P, we incorporate a feedback generative adversarial network (FBGAN) model that combines generative adversarial networks and feedback techniques to effectively tackle the issue of data imbalance. Extensive experiments have shown that DeepB³P outperforms state-of-the-art predictors on the benchmarking dataset. Specifically, DeepB³P exhibits superiority over other BBBP prediction models, achieving nearly 9.09%, 4.55% and 9.41% higher in terms of specificity, accuracy, and Matthew's correlation coefficient respectively. Additionally, interpretable analyses provide valuable insights and downstream analyses for the identification of BBBPs. Furthermore, the FBGAN model trained in this study has the capability to generate novel BBBP-like peptides, which can serve as potential candidates for CNS drug development.