Serial KinderMiner (SKiM) Discovers and Annotates Biomedical Knowledge Using Co-Occurrence and Transformer Models.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Literature-Based Discovery

Background:

• Biomedical research faces information overload with over 34 million PubMed articles.
• Efficient and interpretable tools are crucial for researchers to identify associations between biomedical concepts.
• Literature-based discovery (LBD) aims to uncover hidden connections within isolated scientific literature, often through A-B-C relationships.

Approach:

• Serial KinderMiner (SKiM) is an LBD algorithm designed to find statistically significant A-B-C linkages.
• SKiM addresses limitations of existing LBD tools by providing a functional web interface, identifying relationship types, allowing user-defined concept lists, and querying thousands of terms.
• The algorithm is generalized for any biomedical domain and offers improvements over tools limited to specific areas like cancer research.

Key Points:

• SKiM demonstrates its utility in discovering A-B-C linkages through experiments in classic LBD, drug repurposing, and cancer-related associations.
• A knowledge graph, built using transformer machine-learning models, supplements SKiM to help interpret discovered relationships.
• An open-source web interface (https://skim.morgridge.org) provides intuitive access to SKiM's capabilities, including comprehensive lists of drugs, diseases, phenotypes, and symptoms.

Conclusions:

• SKiM is a versatile algorithm for LBD, enabling relationship discovery between user-defined concepts across any domain.
• It supports large-scale searches involving thousands of terms, surpassing the limitations of many existing LBD tools.
• SKiM enhances relationship identification by providing type labels derived from its knowledge graph, moving beyond simple existence detection.