ProtPhage: a deep learning framework for phage viral protein identification and functional annotation.

Area of Science:

• Computational biology
• Virology
• Genomics

Background:

• Antibiotic-resistant pathogens pose a global health threat.
• Bacteriophages (phages) offer a viable alternative to antibiotics.
• Identifying phage viral proteins (PVPs) is crucial for understanding phage-host interactions but is challenging due to sequence diversity and data limitations.

Purpose of the Study:

• To develop a novel framework, ProtPhage, for accurate identification and functional annotation of phage viral proteins (PVPs).
• To address challenges in PVP prediction, including high sequence diversity and class imbalance.
• To improve the prediction of minority classes, such as minor capsid proteins.

Main Methods:

• Utilized the ProtT5 protein language model for enhanced sequence representation.
• Incorporated an asymmetric loss function to effectively handle class imbalance in datasets.
• Evaluated ProtPhage performance against existing methods using metrics like accuracy, precision, recall, and F1 score.

Main Results:

• ProtPhage significantly improved the prediction of the "minor capsid" class, achieving a 33.07% higher F1 score than the best existing model.
• Demonstrated superior performance across multiple evaluation metrics compared to current state-of-the-art methods.
• Successfully applied ProtPhage to a case study on the Mycobacterium phage PDRPxv genome, validating its practical utility.

Conclusions:

• ProtPhage establishes a new benchmark for PVP identification and annotation in computational phage biology.
• The framework shows potential for broader applications, including phage-host prediction.
• Advanced deep learning techniques, like ProtT5 and asymmetric loss, are effective for analyzing complex biological data.