Peptriever: a Bi-Encoder approach for large-scale protein-peptide binding search

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Peptide therapeutics require precise peptide-receptor interactions.
• Identifying potential protein targets for a given peptide is computationally challenging with existing methods.

Purpose of the Study:

• To develop a novel computational method for rapidly identifying and ranking potential protein targets for specific peptides.
• To improve the prediction accuracy of peptide-protein interactions compared to existing sequence- and structure-based approaches.

Main Methods:

• Utilized a transformers-based protein embedding scheme.
• Developed a method for quick identification and ranking of millions of interacting proteins.

Main Results:

• The proposed approach significantly outperforms existing sequence- and structure-based methods.
• Achieved a mean AUC-ROC and AUC-PR of 0.73, demonstrating high predictive performance.

Conclusions:

• The transformers-based embedding scheme offers an efficient and accurate solution for predicting peptide-protein interactions.
• This method facilitates the identification of potential therapeutic targets for peptide-based drugs.