Link prediction in a MeSH co-occurrence network: preliminary results

Area of Science:

• Biomedical Informatics
• Computational Biology
• Scientific Literature Analysis

Background:

• Literature-based discovery (LBD) aims to automatically identify implicit relationships within scientific texts.
• Co-occurrence networks of biomedical concepts are a common approach in LBD.
• Existing LBD methods can be augmented by predicting missing links in these networks.

Purpose of the Study:

• To propose and evaluate a novel methodology for link prediction in co-occurring Medical Subject Headings (MeSH) networks.
• To enhance existing literature-based discovery techniques through improved identification of implicit biomedical connections.

Main Methods:

• Constructed a network of co-occurring Medical Subject Headings (MeSH) from scientific literature.
• Applied link prediction algorithms, specifically Jaccard and Adamic-Adar similarity measures.
• Evaluated prediction performance using the area under the ROC curve (AUC).

Main Results:

• Achieved high prediction performance for identifying implicit links in the MeSH network.
• The Adamic-Adar similarity measure yielded an AUC of 0.82.
• The Jaccard similarity measure achieved an AUC of 0.78, demonstrating robust performance.

Conclusions:

• The proposed link prediction methodology effectively identifies implicit connections in biomedical literature networks.
• This approach complements existing LBD methods, offering a valuable tool for researchers.
• The high AUC values indicate the potential of network link prediction for advancing biomedical knowledge discovery.