CoCoScore: context-aware co-occurrence scoring for text mining applications using distant supervision.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• Information extraction from scientific literature is crucial for identifying relationships between biomedical entities.
• Existing methods often rely on single-sentence co-mentions or ignore textual context in corpus-wide co-occurrence analysis.
• There is a need for advanced approaches that integrate both corpus-level statistics and contextual information for more accurate relation extraction.

Purpose of the Study:

• To develop and evaluate a novel corpus-wide co-occurrence scoring approach for relation extraction.
• To enhance the accuracy of identifying associations between biomedical entities by considering textual context.
• To provide a versatile text mining tool for uncovering pairwise associations.

Main Methods:

• Proposed CoCoScore, a method that scores the certainty of associations based on contextual co-occurrences across a corpus.
• Employed distant supervision using a gold-standard set of known associations for training.
• Labeled co-mentions as positive or negative based on their presence or absence in the gold standard, avoiding manual annotation.

Main Results:

• CoCoScore demonstrated superior performance compared to existing methods in identifying human disease-gene and tissue-gene associations.
• The approach successfully identified physical and functional protein-protein associations across different species.
• The method proves effective in uncovering pairwise associations through co-occurrence mining.

Conclusions:

• CoCoScore offers a significant advancement in biomedical relation extraction by effectively integrating textual context with corpus-wide co-occurrence data.
• The tool is versatile and applicable to various biomedical association mining tasks and potentially beyond.
• The availability of CoCoScore facilitates further research and application in the field.