ACPPfel: Explainable deep ensemble learning for anticancer peptides prediction based on feature optimization.

Area of Science:

• Biotechnology
• Computational Biology
• Oncology

Background:

• Cancer remains a leading cause of global mortality, with conventional treatments posing risks to vital organ function.
• Anticancer peptides (ACPs) show promise as targeted cancer therapies due to their specificity and lower toxicity.
• Identifying effective ACPs via traditional wet-lab methods is labor-intensive and time-consuming.

Purpose of the Study:

• To develop a computational framework for predicting anticancer peptides (ACPs).
• To accelerate the identification of potential ACPs for cancer treatment.
• To provide a user-friendly tool for researchers in the field of anticancer peptide discovery.

Main Methods:

• A deep ensemble learning model was constructed for ACP prediction.
• Feature selection and dimensionality reduction techniques were integrated into the model training.
• The model's performance was validated using four distinct datasets.
• Interpretability analysis was performed to identify key predictive features.

Main Results:

• The deep ensemble model achieved high predictive accuracy, reaching 98.53% accuracy and an AUC of 0.9972 on the ACPfel dataset.
• The framework demonstrated improved performance across multiple datasets.
• A web server was developed to facilitate accessible ACP prediction for the research community.

Conclusions:

• The deep ensemble learning approach provides an efficient and accurate method for predicting anticancer peptides.
• This computational strategy significantly reduces the time and resources required for ACP identification.
• The developed web server offers a valuable resource for advancing anticancer peptide research and drug discovery.