Contrastive learning for enhancing feature extraction in anticancer peptides.

Area of Science:

• Oncology
• Computational Biology
• Biotechnology

Background:

• Cancer remains a leading global cause of death, necessitating novel therapeutic strategies.
• Anticancer peptides (ACPs) show promise, but their screening is resource-intensive.
• In silico prediction tools offer an efficient alternative for identifying potential ACPs.

Purpose of the Study:

• To develop and evaluate a deep learning model for in silico screening of anticancer peptides (ACPs).
• To enhance ACP prediction accuracy using contrastive learning and dual encoders.
• To assess the model's performance against existing state-of-the-art methods on benchmark datasets.

Main Methods:

• A deep learning model was designed to predict ACPs based solely on peptide sequences.
• Contrastive learning techniques were implemented to improve model performance.
• Two independent encoders replaced traditional data augmentation methods within the contrastive learning framework.

Main Results:

• The proposed model demonstrated superior performance on five out of six benchmark datasets.
• Contrastive learning significantly outperformed models trained only on binary classification loss.
• The use of dual encoders proved effective, potentially replacing data augmentation.

Conclusions:

• The developed deep learning model offers an efficient and accurate method for screening anticancer peptides.
• Advanced computational approaches like contrastive learning can significantly boost the predictive power of ACP identification tools.
• These advancements hold promise for accelerating the development of peptide-based cancer therapeutics and improving patient outcomes.