Harnessing Sequence Embedding and Ensemble Learning to Identify Antifungal Peptides with Low Hemolytic Risk.

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Bioinformatics
• Drug Discovery and Development

Background:

• Rising prevalence of fungal infections and antifungal resistance necessitates novel therapeutic strategies.
• Antifungal peptides (AFPs) show promise as alternatives due to diverse mechanisms and low resistance potential.
• Systematic discovery and prioritization of AFPs are crucial for developing new antifungal treatments.

Purpose of the Study:

• To develop a computational framework, AI4AFP, for predicting antifungal potential from peptide sequences.
• To integrate sequence-based prediction with experimental validation for efficacy and toxicity assessment.
• To provide an accessible platform for the discovery and prioritization of novel antifungal peptides.

Main Methods:

• Developed AI4AFP, a machine learning framework combining multiple sequence encoding strategies (ProtBERT-BFD, PC6, Doc2Vec) and learning algorithms (random forests, SVM, CNNs, BERT).
• Constructed a seven-model ensemble for predicting general antifungal potential from peptide sequences.
• Developed a hemolysis classifier incorporating peptide sequence and concentration to model dose-dependent toxicity, validated by MHC10 determination.

Main Results:

• The AI4AFP ensemble achieved high performance (0.94 accuracy, 0.89 MCC) on an independent test set, outperforming existing methods.
• Experimental validation showed high-scoring AFPs had context-dependent activity against fungal pathogens like Candida albicans and Cryptococcus neoformans.
• The hemolysis classifier effectively modeled dose-dependent toxicity, providing a safety reference alongside antifungal activity predictions.

Conclusions:

• AI4AFP is a robust tool for predicting general antifungal potential and prioritizing AFP candidates.
• Experimental validation is essential to confirm species-specific efficacy and interpret toxicity in context.
• The AI4AFP web server provides a valuable resource for discovering and prioritizing safe and effective antifungal peptides.