PhaVIP

PhaVIP classifies phage virion proteins (PVPs) from protein sequences using chaos game representation and a Vision Transformer to enable structural protein annotation and downstream phage taxonomy and host prediction.

Key Features:

• Chaos Game Representation (CGR): Transforms protein amino acid sequences into image-like representations for image-based analysis.
• Vision Transformer (ViT): Applies a Vision Transformer image-classification model to CGR images to perform feature learning.
• Feature extraction: Captures both local and global sequence-derived features from CGR images for classification.
• Classification tasks: Performs binary PVP versus non-PVP discrimination and multi-class annotation of PVP types such as capsid, tail, major tail, and baseplate components.
• Benchmarking and performance: Evaluated on multiple datasets and benchmarked against alternative tools with reported improvements in accuracy and speed.
• Input sources: Applicable to protein sequences derived from high-throughput sequencing datasets.

Scientific Applications:

• Structural protein annotation: Identification and annotation of virion structural proteins including capsid, tail, major tail, and baseplate proteins.
• Phage taxonomy classification: Use of classified PVPs to improve phage taxonomic assignments.
• Phage host prediction: Use of PVP annotations to support prediction of phage host ranges.
• Experimental prioritization: Prioritization of candidate virion proteins from large sequencing datasets for experimental validation.

Methodology:

Protein sequences are converted to chaos game representation (CGR) images, which are input to a Vision Transformer (ViT) that extracts local and global features for binary PVP/non-PVP classification and multi-class PVP type annotation, with models benchmarked on multiple datasets against alternative tools.