FunVFPred: Predicting fungal virulence factors using a unified representation learning model

Area of Science:

• Mycology
• Computational Biology
• Infectious Diseases

Background:

• Fungal virulence factors (VFs) are crucial for pathogenic fungi to colonize hosts, invade tissues, modulate immune responses, and cause disease.
• Identifying VFs is essential for understanding fungal-host interactions and developing targeted antifungal strategies.
• Current methods for VF identification can be labor-intensive and require extensive experimental validation.

Purpose of the Study:

• To develop and validate a machine learning framework, FunVFPred, for predicting fungal virulence factors (VFs) from protein sequences.
• To assess the framework's accuracy and generalizability across different pathogenic fungal species.
• To provide a computational resource for prioritizing candidate VFs for experimental investigation.

Main Methods:

• Developed FunVFPred, a machine learning framework utilizing UniRep protein sequence embeddings and multiple classifiers.
• Trained and evaluated models using experimentally validated VFs from human-pathogenic *Candida* species.
• Validated the framework on an independent dataset of *Aspergillus fumigatus* VFs.

Main Results:

• The random forest model achieved 73.4% accuracy and an MCC of 0.47 in cross-validation on *Candida* species.
• On an independent *Aspergillus fumigatus* dataset, FunVFPred demonstrated 85.7% accuracy and an MCC of 0.71.
• These results indicate the potential for cross-species prediction and prioritization of fungal virulence-associated proteins.

Conclusions:

• FunVFPred is an effective machine learning tool for predicting fungal virulence factors.
• The framework shows promise for cross-species application in identifying virulence-associated proteins.
• FunVFPred offers an accessible computational resource to advance the study of fungal pathogenicity and guide experimental validation efforts.