Discovering discovery patterns with Predication-based Semantic Indexing.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Drug Discovery

Background:

• Literature-based discovery (LBD) aims to identify novel therapeutic relationships from existing scientific literature.
• Traditional LBD methods often struggle with scalability and efficiency in processing vast amounts of biomedical text.
• Semantic predications extracted by systems like SemRep offer a structured way to represent relationships within the literature.

Purpose of the Study:

• To develop and evaluate a novel hyperdimensional computing approach for literature-based discovery.
• To identify and utilize "discovery patterns"—sequences of semantic relationships—to link pharmaceutical substances to diseases.
• To enhance the efficiency and accuracy of finding therapeutic relationships for drug repurposing.

Main Methods:

• Utilized hyperdimensional computing methods for identifying therapeutically relevant connections.
• Developed Predication-based Semantic Indexing (PSI) to identify "discovery patterns" from SemRep-extracted semantic predications.
• Applied geometric operators in PSI space for rapid inference and to direct searches for therapeutic relationships.

Main Results:

• Successfully rediscovered manually identified discovery patterns.
• Discovered a set of previously unrecognized therapeutic discovery patterns.
• PSI-based search demonstrated superior recovery of therapeutic relationships compared to distributional statistics models.
• Demonstrated the utility of efficient approximate inference in geometric space for identifying therapeutic relationships.

Conclusions:

• Hyperdimensional computing and PSI offer an efficient and effective approach for literature-based drug discovery.
• The discovery pattern approach is valuable for identifying novel therapeutic uses of existing drugs (drug repurposing).
• This method holds significant promise for advancing drug repurposing efforts in the pharmaceutical industry.