DeePhafier: a phage lifestyle classifier using a multilayer self-attention neural network combining protein...

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Bacteriophages (phages) are viruses infecting bacteria, crucial for microbiome dynamics.
• Phages are classified as virulent or temperate, impacting phage-host interactions.
• Existing classification methods lack accuracy due to limited feature considerations.

Purpose of the Study:

• To introduce DeePhafier, a novel computational method for improved phage lifestyle classification.
• To enhance the understanding of phage-host interactions through accurate classification.
• To provide a more robust tool for analyzing phage genomes.

Main Methods:

• Development of DeePhafier utilizing multilayer self-attention neural networks and a global self-attention neural network.
• Integration of protein features derived from Position Specific Scoring Matrix (PSSM).
• Evaluation using five-fold cross-validation on phage sequences.

Main Results:

• DeePhafier achieved a high classification accuracy of 87.54% for sequences longer than 2000bp.
• The method demonstrated superior performance compared to two benchmark classification approaches.
• The combined use of sequence and protein features significantly improved classification accuracy.

Conclusions:

• DeePhafier offers a significant advancement in classifying bacteriophage lifestyles.
• The computational approach effectively leverages deep learning and protein features for enhanced accuracy.
• Accurate phage lifestyle classification is vital for microbiome research and understanding phage biology.