NetQuilt

NetQuilt integrates homology and network similarity to predict protein functions across multiple species using multispecies protein-protein interaction (PPI) networks.

Key Features:

• Homology-informed network integration: Combines sequence data with network-based information across species to provide cellular-context-aware features for function prediction.
• IsoRank-based cross-species alignment: Uses IsoRank to compute similarity scores between PPI networks from different organisms.
• Multispecies meta-network construction: Aggregates IsoRank-derived similarities into a comprehensive meta-network profile spanning multiple species.
• Maxout neural network model: Trains a maxout neural network using the integrated multispecies meta-network as input.
• Gene Ontology supervision: Uses Gene Ontology (GO) terms as target labels for supervised function prediction.
• Robustness to incomplete PPI data: Leverages the multispecies meta-network to enable predictions for species with missing or incomplete PPI networks.
• Input data and databases: Operates on sequences, PPI networks, and GO annotations, with datasets obtainable from STRING.
• Comparative performance: Demonstrates improved prediction accuracy relative to single-species network-based methods, deep learning sequence-based approaches, and BLAST-based annotation transfer.

Scientific Applications:

• Cross-species functional annotation: Assigns GO terms to proteins by leveraging multispecies interaction context and homology signals.
• Study of functional conservation and divergence: Enables analysis of conserved and divergent functional signals across species using integrated network similarity.
• Function prediction for organisms with sparse data: Produces predictions for species that lack complete PPI networks by borrowing information from related species.
• Method benchmarking: Facilitates comparison of network-based, sequence-based, and homology-based annotation approaches.

Methodology:

NetQuilt integrates sequence data and PPI networks across species, computes cross-species network similarity with IsoRank to build a multispecies meta-network, and trains a maxout neural network using Gene Ontology (GO) terms as labels.