cross-attention PHV

cross-attention PHV predicts protein-protein interactions (PPIs) between human and viral proteins to support analysis of viral infection mechanisms, host immune responses, and potential therapeutic targets.

Key Features:

• Cross-Attention Mechanism: Implements a cross-attention mechanism to focus on relevant features within protein sequences and enhance prediction accuracy and generalization.
• One-Dimensional Convolutional Neural Network (1D-CNN): Integrates 1D-CNNs to reduce computational cost and enable processing of long protein sequences.
• word2vec-Generated Feature Matrices: Uses word2vec to convert protein sequences into feature matrices for downstream neural network processing.
• Long-Sequence Support: Capable of processing protein sequences up to 9000 amino acid residues.
• Performance: Demonstrated superior performance on benchmark datasets, achieving area under the curve (AUC) values greater than 0.95 for human–SARS-CoV-2 PPIs.

Scientific Applications:

• Host-virus interaction mapping: Predicts candidate human-virus PPIs to inform studies of viral infection mechanisms and host immune responses.
• Therapeutic target identification: Identifies putative host or viral proteins involved in interactions that may serve as therapeutic targets.
• Emerging pathogen analysis and experimental complementarity: Supports exploration of interactions for emerging viruses and complements experimental approaches such as mass spectrometry-based proteomics and yeast two-hybrid assays.

Methodology:

Protein sequences are converted into word2vec-generated feature matrices and processed by an integrated 1D-CNN and cross-attention neural network architecture to predict human–virus PPIs.