HiG2Vec

HiG2Vec embeds Gene Ontology (GO) terms, gene ontology annotations (GOA), and genes into a Poincaré ball to capture hierarchical relationships and produce vector representations for semantic similarity and interaction analyses.

Key Features:

• Hierarchical Representation: Uses Poincaré embeddings with a two-step procedure (GO embedding then gene embedding) to represent hierarchical relationships.
• Semantic Similarity Measurement: Produces vector representations that improve semantic similarity assessment between GO terms and genes compared to Word2Vec-based Euclidean embeddings.
• Preservation of Latent Hierarchies: Embeds entities in hyperbolic space to preserve latent hierarchical information often lost in Euclidean methods.
• Numeric Vectors for Deep Learning: Outputs numeric vector representations suitable for integration into deep learning workflows.
• Gene Interaction Prediction: Enables prediction of gene or gene product interactions with performance comparable to or exceeding prior methods.

Scientific Applications:

• Deep Learning Integration: Provides vectorized GO and gene representations for use as input features in deep learning models.
• Gene Interaction Prediction: Supports computational prediction of gene/gene product interactions using embeddings that capture semantic and hierarchical context.

Methodology:

Two-step embedding: first embed GO terms into the Poincaré ball to capture hierarchical relationships, then embed genes in relation to these GO term embeddings.