GeneWalk

GeneWalk identifies condition-specific gene functions by assembling experiment-specific gene regulatory networks and using representation learning with random walks to score Gene Ontology (GO) annotations relative to experimental conditions.

Key Features:

• Condition-specific gene regulatory network assembly: Automatically constructs experiment- or condition-specific gene regulatory networks reflecting interactions under particular experimental settings.
• Representation learning of genes and GO terms: Learns vector representations for individual genes and their annotated Gene Ontology (GO) terms and quantifies similarities between them.
• Random walks on experiment-specific networks: Uses random walks on the assembled networks to extract contextual features for genes.
• Comparison with randomized networks: Compares node similarities derived from real and randomized networks to assess the significance of functional annotations.
• Annotation significance scoring: Produces annotation significance scores tailored to reflect experimental conditions.
• Gene- and condition-specific functional analysis: Transforms lists of gene hits into data-driven, gene-level functional hypotheses specific to the experimental context.
• Complement to GO enrichment: Provides gene-level functional insights that complement traditional GO enrichment, which typically reports gene set–level results.

Scientific Applications:

• Functional genomics: Analyze gene functions in high-throughput functional genomics experiments with context specificity.
• Context-specific annotation: Identify Gene Ontology (GO) annotations that are relevant to particular experimental conditions.
• Hypothesis generation: Generate data-driven hypotheses at the individual gene level from lists of gene hits.
• Network-based functional interpretation: Interpret gene-level roles and interactions using experiment-specific regulatory networks.

Methodology:

GeneWalk assembles condition- or experiment-specific gene regulatory networks, performs random walks on these networks to obtain representation learning–based vector embeddings for genes and annotated GO terms, computes similarities between gene and GO vectors, and compares similarities from real versus randomized networks to produce annotation significance scores.