MeSHHeading2vec: a new method for representing MeSH headings as vectors based on graph embedding algorithm.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Data Science

Background:

• Medical Subject Headings (MeSH) terms are crucial for organizing biomedical information but are abstract and difficult to quantify.
• Developing discriminative vector representations for MeSH terms can significantly enhance computational prediction models in medicine and biology.

Purpose of the Study:

• To convert the MeSH hierarchical structure into a relationship network.
• To apply graph embedding algorithms for generating vector representations of MeSH terms.
• To evaluate the effectiveness of these vector representations in downstream computational tasks.

Main Methods:

• Constructed a MeSH relationship network using tree numbers.
• Applied five graph embedding algorithms: DeepWalk, LINE, SDNE, LAP, and HOPE.
• Evaluated performance through node classification and relationship prediction tasks.

Main Results:

• Graph embedding algorithms effectively represent MeSH headings as vectors.
• These vector representations can be used independently or as supplementary information to improve existing vector capabilities.
• The method demonstrates utility for computational models involving diseases, drugs, and microbes.

Conclusions:

• Network-based graph embedding provides a powerful method for representing abstract MeSH terms.
• This approach enhances the performance of computational prediction models in biomedical research.
• The findings offer a novel perspective for representing terms within complex networks.