Knowledge graph completion

As explained in Graph Algorithms for Data Science

Knowledge graph completion is the task of inferring missing entity relationships and their relationship types from an incomplete knowledge graph.

Knowledge graph completion infers missing links and their relationship types in a graph containing multiple node and relationship types. It is a multiclass form of link prediction: link prediction forecasts future links, while completion fills missing ones. Unlike homogeneous-graph features such as common-neighbor counts, completion distinguishes entities and relationship types. section 11.0 section 11.2

A knowledge graph embedding model encodes nodes and relationships in an embedding space. It learns representations in which the head and relationship of an existing triple align with its tail, then scores possible missing triples. section 11.4

A practical completion workflow follows these steps:

• Start with a rich, heterogeneous graph containing entities and typed relationships. section 11.2
• Load relationship data, such as a CSV file, into a pandas dataframe when needed. section 11.2.1
• Train a knowledge graph embedding model to represent nodes and relationships. section 11.2
• Rank candidate links and their types, such as new TREATS relationships for a known drug. section 11.2.1
• Validate promising predictions with literature or domain experts before clinical evaluation. section 11.2.5 section 11.4

Drug repurposing with Hetionet

Use the Hetionet biomedical graph to predict new TREATS relationships from acetylsalicylic acid to diseases. Its full dataset contains 47,031 nodes across 11 types and 2,250,197 relationships across 24 types; the workflow uses a smaller subset. PyKEEN supplies embedding models and prediction methods, while supporting paths between aspirin and a candidate disease help experts assess the result. Knowledge graph completion prioritizes candidates; it does not replace clinical trials. section 11.2 section 11.2.1 section 11.2.5 section 11.4

Figure 11.11 Drug repurposing workflow

Use the drug-repurposing workflow diagram to connect graph data, embedding-model training, candidate prediction, and evaluation. section 11.2

How Graph Algorithms for Data Science covers it

How deep it goes

Explains heterogeneous graphs, triples, embeddings, prediction, and validation, but stops short of optimization details and clinical approval.

Examples

Demonstrates Hetionet, aspirin, TREATS relationships, PyKEEN, pandas, and short paths connecting aspirin with pancreatic cancer.

What it assumes

Assumes basic graph and data-handling familiarity while explaining typed triples, embeddings, and heterogeneous graphs from the workflow.

What you can do after

Apply it via PyKEEN to rank candidate relationships, then diagnose predictions through literature, graph paths, and expert review.

How it connects

Connects heterogeneous graphs, knowledge graph embeddings, multiclass link prediction, drug repurposing, and clinical trials.

liveBookmanning.com

Read Graph Algorithms for Data Science on liveBook Get the book at manning.com

Related concepts

• knowledge graph embeddinguses · Knowledge graph embedding is a technique that represents entities and relationships as low-dimensional vectors while preserving patterns expressed by valid triples.
• link predictionkind of · Link prediction is a graph-learning task that estimates whether a connection exists or may occur between nodes, often by scoring candidate links.
• embeddinguses · A numerical vector representation of an item or input that encodes features or relationships useful for computational tasks.
• Hetionetuses · Hetionet is a biomedical knowledge graph integrating entities such as diseases, genes, pathways, and compounds through typed relationships.
• drug repurposingused by · Drug repurposing is the investigation or use of an existing drug for a therapeutic indication different from its original use.

This explainer was generated by Manning's knowledge graph from the text of Graph Algorithms for Data Science. The section links point to the parts of the book it draws on.