PhANNs, a fast and accurate tool and web server to classify phage structural proteins

Area of Science:

• Computational Biology
• Bioinformatics
• Virology

Background:

• A significant portion (50-90%) of genes in bacteriophage genomes and metagenomic data lack functional assignment.
• Structural protein genes are abundant in phage genomes but are highly divergent and challenging to identify using traditional homology methods.

Purpose of the Study:

• To develop a novel computational approach for classifying bacteriophage open reading frames (ORFs) into functional categories.
• To enhance the functional annotation of phage proteins, aiding in understanding phage roles and potential therapeutic applications.

Main Methods:

• Development of Phage Artificial Neural Networks (PhANNs), a tool utilizing Artificial Neural Network ensembles.
• Creation of a curated database of 538,213 phage protein sequences.
• Implementation of a novel clustering method for training set creation, ensuring sequence diversity and minimizing homology between sets.

Main Results:

• PhANNs achieved a test F1-score of 0.875 and 96.2% test accuracy in classifying phage proteins.
• The tool categorizes proteins into ten major structural classes or an 'other' category.
• PhANNs provides rapid and accurate functional annotation for phage proteins.

Conclusions:

• PhANNs offers a new, effective method for the functional annotation of bacteriophage proteins, particularly structural genes.
• This tool facilitates a deeper understanding of phage biology, ecological contributions, and potential use in phage therapy.
• PhANNs is open-source and accessible via a web server or local installation.