AsyRW

AsyRW predicts protein functions across species by performing asynchronous-random walks on a heterogeneous network that integrates cross-species annotations and homology relationships, using a gravity-like theory to quantify node-specific walk lengths for association with Gene Ontology (GO) terms.

Key Features:

• Heterogeneous network construction: Constructs a heterogeneous network integrating multiple functional association networks derived from diverse biological data sources.
• Cross-species homology and annotations: Incorporates established homology relationships between proteins across species, known protein annotations, and Gene Ontology (GO) terms.
• Gravity-like theory for walk length quantification: Employs a gravity-like theory to quantify individual walk lengths for each network node, accounting for intrinsic structures of intra-species and inter-species proteins and GO terms.
• Asynchronous-random walk algorithm: Performs asynchronous-random walks using the quantified individual walk lengths to predict associations between proteins and GO terms.
• Leveraging complementary annotations: Exploits complementary annotations from different species to enhance prediction performance.
• Temporal performance evaluation: Demonstrates improved prediction performance on annotations archived in different years compared to related methods.

Scientific Applications:

• Cross-species protein function prediction: Integrates data across species to predict protein functions that are incompletely annotated within single species.
• Protein–GO association prediction: Assigns Gene Ontology terms to proteins by inferring associations from the heterogeneous network via asynchronous-random walks.
• Annotation transfer and improvement: Enhances annotation accuracy by transferring complementary functional information between species.

Methodology:

Construct a heterogeneous network integrating multiple functional association networks, homology relationships, protein annotations, and GO terms; quantify node-specific walk lengths using a gravity-like theory; and perform asynchronous-random walks with those lengths to predict protein–GO associations.