SiameseCPP: a sequence-based Siamese network to predict cell-penetrating peptides by contrastive learning

Area of Science:

• Biochemistry
• Computational Biology
• Drug Delivery

Background:

• Cell-penetrating peptides (CPPs) are crucial for intracellular delivery of therapeutic molecules.
• Existing machine learning methods for CPP prediction often rely on handcrafted features and lack consideration of peptide similarity.
• There is a need for advanced computational tools to accurately identify CPPs for therapeutic applications.

Purpose of the Study:

• To develop a novel deep learning framework, SiameseCPP, for automated and accurate prediction of cell-penetrating peptides.
• To leverage contrastive learning and Siamese neural networks for enhanced CPP representation learning.
• To evaluate the performance and generalization ability of SiameseCPP compared to existing methods.

Main Methods:

• Developed SiameseCPP, a deep learning framework utilizing a pretrained model and a Siamese neural network (transformer and GRUs).
• Implemented contrastive learning for the first time in a CPP predictive model to learn discriminative representations.
• Conducted comprehensive experiments to compare SiameseCPP against baseline models.

Main Results:

• SiameseCPP significantly outperforms existing baseline models in predicting cell-penetrating peptides.
• The model demonstrates strong generalization ability, achieving good performance on other functional peptide datasets.
• Contrastive learning proved effective for building a robust CPP predictive model.

Conclusions:

• SiameseCPP represents a significant advancement in the automated prediction of cell-penetrating peptides.
• The framework's superior performance and generalization capability highlight its potential for therapeutic development.
• This study validates the efficacy of contrastive learning in the domain of peptide function prediction.