ELP

ELP predicts the likelihood of enzymatic transformations between molecules using graph-based learning to infer biochemical reaction links.

Key Features:

• Graph-Based Modeling: Models enzymatic reactions as a graph derived from the KEGG database.
• Graph Embedding: Learns molecular and enzyme representations via graph embedding that capture node attributes and network connectivity.
• Learning Modes: Explores both transductive and inductive learning models, addressing training with or without test nodes included in the graph.
• Use of Node Attributes: Combines molecular and enzymatic attributes with graph connectivity to inform embeddings and link prediction.
• Performance Metrics: Reports improved Area Under Curve (AUC) performance, outperforming fingerprint-based similarity approaches by 30% in AUC and support vector machines by 8%.
• Comparative Evaluation: Evaluated against rule-based methods for enzymatic link prediction.

Scientific Applications:

• Pathway Map Prediction: Predicts links within pathway maps to aid elucidation of biochemical pathways.
• Reaction Network Reconstruction: Reconstructs edges in reaction networks for gut microbiota phyla including actinobacteria, bacteroidetes, firmicutes, and proteobacteria.
• Visualization Guidance: Illustrates the importance of graph embedding within biochemical networks to guide visualization of complex biological data.

Methodology:

Constructs a reaction graph from KEGG, applies graph embedding to learn node representations from molecular and enzymatic attributes and graph connectivity, and explores transductive and inductive machine learning models for link prediction.