pathway2vec

pathway2vec generates neural embeddings from a three-layer heterogeneous network of compounds, enzymes, and pathways to infer metabolic pathway presence and represent metabolic relationships.

Key Features:

• Heterogeneous Network Embedding: Constructs a three-layer network integrating compounds, enzymes, and pathways and captures intra-layer and inter-layer interactions to represent metabolic relationships.
• Representational Learning Modules: Implements six distinct modules that use neural embedding techniques to transform biological data into a low-dimensional feature space.
• Machine Learning Integration: Infers pathway presence using heuristic or algorithmic machine-learning approaches as an alternative to gene-centric methods.

Scientific Applications:

• Pathway Prediction: Uses neural embeddings to predict metabolic pathways from genomic data and validates predictions against MetaCyc.
• Benchmarking Performance: Evaluates embeddings via node clustering and embedding visualization to benchmark representation quality and prediction outcomes.
• Research Applications: Generates predictive models from genomic data to support studies of cellular function, disease mechanisms, and ecological interactions at the community level.

Methodology:

Builds and analyzes a three-layer network (compounds, enzymes, pathways), learns embeddings in a low-dimensional space, and applies those embeddings to node clustering, embedding visualization, and pathway prediction.