dRiskKB: a large-scale disease-disease risk relationship knowledge base constructed from biomedical text

Area of Science:

• Computational biology
• Bioinformatics
• Medical informatics

Background:

• Discerning genetic contributions to complex diseases requires advanced computational methods.
• Systems approaches to disease relationships are crucial for gene discovery and drug repositioning.
• Lack of large-scale, machine-understandable disease relationship knowledge bases limits phenome-wide studies.

Purpose of the Study:

• To develop a precise, large-scale disease-disease risk relationship knowledge base (dRiskKB).
• To leverage a semi-supervised iterative pattern learning approach for knowledge base construction.
• To facilitate disease etiology research and drug repositioning.

Main Methods:

• Utilized 21,354,075 MEDLINE records.
• Employed a semi-supervised iterative pattern learning approach starting with a seed pattern.
• Extracted disease risk pairs (D1 → D2) and manually evaluated precision.
• Analyzed correlations between disease risk pairs, genes, and drugs.

Main Results:

• Constructed dRiskKB with 34,448 unique D1 → D2 risk pairs among 12,981 diseases.
• Achieved high precision for learned patterns (0.99) and extracted pairs (0.919-0.988).
• Demonstrated algorithm robustness and identified shared genes and drugs for diseases with similar risk profiles.

Conclusions:

• The dRiskKB offers a valuable resource for understanding disease etiology.
• Integration with other datasets can enhance disease understanding and drug repositioning efforts.
• The developed computational approach is effective for building large-scale biomedical knowledge bases.