Optimizations for Computing Relatedness in Biomedical Heterogeneous Information Networks: SemNet 2.0

Area of Science:

• Biomedical Informatics
• Computational Biology
• Knowledge Discovery

Background:

• Literature-based discovery (LBD) is crucial for summarizing information from large text corpuses.
• Existing frameworks like SemNet use knowledge graphs for concept relatedness and ranking.
• Analyzing complex disease etiology and therapeutics requires efficient LBD tools.

Purpose of the Study:

• To enhance the SemNet framework for improved efficacy and efficiency in biomedical LBD.
• To develop SemNet 2.0, a more user-friendly and faster LBD software.
• To optimize knowledge graph query times and rank aggregation algorithms.

Main Methods:

• Replaced Neo4j with a custom Python data structure for knowledge graph management.
• Developed two randomized algorithms to optimize HeteSim metric calculations for metapath similarity.
• Reconstructed the unsupervised learning algorithm for rank aggregation (ULARA) with mathematical proofs and performance guarantees.

Main Results:

• Achieved knowledge graph query time improvements of several orders of magnitude.
• Enhanced ULARA algorithm for broader generalizability beyond SemNet.
• Demonstrated SemNet 2.0's capability for faster, more effective, and user-friendly automated biomedical LBD.

Conclusions:

• SemNet 2.0 offers a significant advancement in automated biomedical LBD.
• The open-source software provides a comprehensive solution for complex data analysis.
• The system facilitates ranking relationships between diseases and comorbidities, exemplified by Alzheimer's disease.