Comparative analysis of tissue-specific anticancer peptide prediction models: ACP-Boost framework.

Area of Science:

• Computational biology and bioinformatics
• Cancer research and therapeutics
• Machine learning applications in drug discovery

Background:

• Conventional cancer treatments face limitations like toxicity and drug resistance.
• Anticancer peptides (ACPs) show promise but require efficient identification methods.
• Existing computational ACP prediction models often lack tissue-specific context.

Purpose of the Study:

• To develop a tissue-aware machine learning framework, ACP-Boost, for classifying anticancer peptides.
• To enable tissue-specific ACP prediction across nine distinct cancer-related tissues.
• To explore the influence of biological heterogeneity on peptide prediction accuracy.

Main Methods:

• Integrated and preprocessed experimentally validated peptide data from CancerPPD2 and DCTPep.
• Encoded peptide sequences into 473-dimensional feature vectors using composition and physicochemical properties.
• Employed a one-versus-rest classification strategy with XGBoost, comparing it against other algorithms.

Main Results:

• XGBoost demonstrated the most stable performance for tissue-specific ACP classification.
• Peptide sequence descriptors contain significant tissue-associated signals, though separability varies by cancer type.
• Feature importance analysis revealed both shared charge properties and tissue-specific descriptors are crucial for discrimination.

Conclusions:

• ACP-Boost provides a robust computational framework for tissue-specific anticancer peptide classification.
• Incorporating biological heterogeneity is vital for improving the accuracy of peptide prediction models.
• The findings support the development of more targeted and effective peptide-based cancer therapies.