PepFormer: End-to-End Transformer-Based Siamese Network to Predict and Enhance Peptide Detectability Based on...

Area of Science:

• Proteomics
• Computational Biology
• Bioinformatics

Background:

• Peptide detectability is crucial for shotgun proteomics.
• Existing computational prediction methods have limitations.
• Accurate prediction requires advanced modeling techniques.

Purpose of the Study:

• To develop a novel, highly accurate computational model for predicting peptide detectability.
• To leverage deep learning, specifically Siamese networks and Transformer architectures, for this task.
• To improve the generalization ability and interpretability of peptide detectability prediction.

Main Methods:

• Developed PepFormer, an end-to-end Siamese network with a hybrid Transformer and gated recurrent unit architecture.
• Utilized contrastive learning and a novel loss function for improved model generalization.
• Trained and evaluated the model on benchmark datasets from *Homo sapiens* and *Mus musculus*.

Main Results:

• PepFormer significantly outperforms state-of-the-art methods on benchmark datasets.
• The model demonstrates strong cross-species transfer learning and adaptability.
• Visualization of embedded representations confirms the model captures high-latent discriminative information.

Conclusions:

• PepFormer offers a robust and accurate solution for predicting peptide detectability.
• The model's architecture and training strategy enhance predictive performance and generalization.
• PepFormer holds significant potential for advancing proteomics studies across different species.