A study on large-scale disease causality discovery from biomedical literature

Area of Science:

• Biomedical Informatics
• Computational Biology
• Medical Natural Language Processing

Background:

• Biomedical semantic relationship extraction is vital for knowledge discovery and AI applications.
• Identifying disease causal relationships aids in understanding pathogenesis and improving treatments.
• Existing tools like SemRep require optimization for accurate disease causality extraction.

Purpose of the Study:

• To optimize the SemRep tool for enhanced accuracy in extracting disease causal relationships.
• To develop a precise semantic predicate vocabulary for disease causality.
• To support automatic extraction of disease causality knowledge from biomedical literature.

Main Methods:

• Constructed a disease causality semantic predicate vocabulary using feature words and SemMedDB data.
• Filtered and evaluated clue words via quantitative comparisons.
• Extracted disease causality pairs using 36 semantic predicates with >80% accuracy.
• Conducted knowledge discovery on extracted disease causality triples (unidirectional, bidirectional, primary, rare).

Main Results:

• Developed a vocabulary of 50 textual predicates with >40% accuracy.
• Utilized 36 predicates for extraction, yielding 259,434 disease causality pairs.
• Identified 176,010 unidirectional and 83,424 bidirectional disease causality triples.
• Discovered primary and rare disease causality types.

Conclusions:

• The proposed method significantly improves SemRep's disease causality extraction accuracy and comprehensiveness.
• Enables automatic extraction of disease causality from large-scale biomedical literature.
• Allows customized and flexible disease causality extraction using a quantified predicate vocabulary.