GeoCTP: Structure-aware Prediction of Multifunctional Cancer Therapy Peptides via Graph Transformer and Contrastive...

Area of Science:

• Bioinformatics and Computational Biology
• Drug Discovery and Development
• Structural Biology

Background:

• Cancer therapy peptides (CTPs) show promise for targeted cancer treatment.
• Traditional experimental screening for CTPs is slow and resource-intensive.
• Developing computational tools is crucial to accelerate CTP discovery.

Purpose of the Study:

• To introduce GeoCTP, a novel geometric deep learning framework for predicting multifunctional CTPs.
• To integrate peptide sequence and 3D structural information for enhanced prediction accuracy.
• To provide a computational tool that aids in identifying potential functional sites on CTPs.

Main Methods:

• Utilized ESMfold for generating peptide 3D structures.
• Employed a Graph Transformer for extracting structure-aware representations.
• Leveraged the ESM-2 language model for semantic feature extraction from sequences.
• Implemented a two-level contrastive learning strategy for feature alignment and discriminability.

Main Results:

• GeoCTP demonstrated superior performance compared to existing state-of-the-art peptide function prediction methods.
• The framework successfully predicted multifunctional CTPs with high accuracy.
• Identified high-attention regions within peptides, suggesting potential functional sites.

Conclusions:

• GeoCTP is the first predictive tool specifically designed for multifunctional CTPs.
• The study highlights the potential of integrating structural and sequence data for CTP prediction.
• GeoCTP offers significant implications for bioinformatics and precision medicine by accelerating CTP discovery.