Cancer therapy peptides (CTPs), as multifunctional peptides, possess the ability to target cancer cells or related proteins, exhibiting significant therapeutic potential. However, traditional experimental screening methods are time-consuming and labor-intensive, limiting the pace of discovery. To address this challenge, we propose GeoCTP, a geometric deep learning framework that integrates both sequence and structural information for accurate CTP prediction. Specifically, GeoCTP employs ESMfold to generate peptide 3D structures and utilizes a Graph Transformer to extract structure-aware representations. For semantic feature extraction from sequences, the ESM-2 language model is adopted. Additionally, a two-level contrastive learning strategy is employed to enhance feature alignment across modalities and improve inter-class discriminability. Experimental results indicate that GeoCTP outperforms state-of-the-art peptide function prediction methods. As the first predictive tool specifically designed for multifunctional CTPs, GeoCTP not only achieves strong classification performance but also identifies high-attention regions indicative of potential functional sites. Overall, this study highlights the potential of CTP functional prediction and its implications for bioinformatics in precision medicine.
Guan et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: