An automatic hypothesis generation for plausible linkage between xanthium and diabetes

Area of Science:

• Biomedical Informatics
• Pharmacology
• Computational Biology

Background:

• Text mining and literature-based discovery accelerate drug development by generating novel hypotheses.
• Previous research has utilized these methods to reduce experimental time and costs for identifying drug candidates.

Purpose of the Study:

• To apply a closed discovery approach and knowledge graph construction for identifying potential natural product drug candidates for diabetes.
• To leverage text mining tools and graph embeddings for hypothesis generation and ranking of therapeutic compounds.

Main Methods:

• Utilized Swanson's ABC model for literature collection on Xanthium compounds and diabetes.
• Employed the Public Knowledge Discovery Engine for Java (PKDE4J) to extract biomedical entities and relations.
• Constructed a knowledge graph, generated paths between Xanthium compounds and diabetes, and ranked them using graph embeddings.

Main Results:

• Identified 35 out of 36 Xanthium compounds with direct paths to diabetes-related nodes.
• Ranked over 2.7 million paths between compounds and diabetes types (type 1, type 2, diabetes mellitus).
• Highlighted adenosine, choline, beta-sitosterol, rhamnose, and scopoletin as top candidates for diabetes drug development.

Conclusions:

• The developed framework effectively generates hypotheses for drug discovery by uncovering biological linkages.
• The study demonstrates the potential of natural products, specifically identified Xanthium compounds, for diabetes treatment.
• The integration of text mining, knowledge graphs, and graph embeddings provides a robust method for identifying novel therapeutic candidates.