Disease phenotype synonymous prediction through network representation learning from PubMed database.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Medical Terminology

Background:

• Phenotype concept mapping between disparate terminology databases is hindered by independent development, leading to incomplete existing maps.
• Manual mapping is labor-intensive and time-consuming, necessitating automated solutions for efficient and comprehensive synonym identification.

Purpose of the Study:

• To develop and evaluate an automated method for predicting synonymous relationships between phenotype concepts from different terminology databases.
• To introduce a classifier-based phenotype mapping prediction model (CPM) that leverages network semantic representations.

Main Methods:

• The CPM model utilizes network semantic representations of phenotypes as input.
• Binary classifiers are trained with a voting strategy to predict synonymous relationships.
• Performance was compared against a similarity-based phenotype mapping prediction model (SPM) and a previous SSDTM method.

Main Results:

• The CPM model, using N2V-TFIDF network representation and a majority voting strategy (MV), achieved an accuracy of 0.943.
• This represents a 15.4% improvement over the SPM using cosine similarity (0.789) and a 23.8% improvement over the SSDTM method (0.724).

Conclusions:

• The proposed classifier-based phenotype mapping prediction model (CPM) demonstrates superior performance in identifying synonymous phenotype concepts.
• Automated prediction of phenotype synonymy is crucial for improving data integration and analysis in biomedical research.