CACPP: A Contrastive Learning-Based Siamese Network to Identify Anticancer Peptides Based on Sequence Only.

Area of Science:

• Biochemistry
• Computational Biology
• Oncology

Background:

• Anticancer peptides (ACPs) show promise in cancer therapy due to their efficacy and safety.
• Experimental identification of ACPs is costly and time-consuming.
• Traditional machine learning methods for ACP prediction suffer from low performance due to reliance on hand-crafted features.

Purpose of the Study:

• To develop a novel deep learning framework, CACPP, for accurate prediction of anticancer peptides.
• To improve upon existing ACP prediction methods by leveraging advanced deep learning techniques.

Main Methods:

• Proposed CACPP, a deep learning framework integrating Convolutional Neural Network (CNN) and contrastive learning.
• Utilized TextCNN to extract high-latent features directly from peptide sequences.
• Employed a contrastive learning module to enhance feature representation distinguishability.

Main Results:

• CACPP demonstrated superior performance compared to state-of-the-art methods on benchmark datasets.
• Feature visualization confirmed the model's robust classification ability.
• Analysis explored dataset construction impact and performance with verified negative samples.

Conclusions:

• CACPP offers a highly accurate and efficient method for anticancer peptide prediction.
• The deep learning approach overcomes limitations of traditional feature engineering.
• This framework advances the identification of novel therapeutic peptides for cancer treatment.