Semantic Disease Gene Embeddings (SmuDGE): phenotype-based disease gene prioritization without phenotypes

Area of Science:

• Computational biology
• Bioinformatics
• Genetics

Background:

• Computational disease gene prioritization methods often rely on gene-phenotype associations.
• Existing methods face limitations due to incomplete gene-phenotype knowledge in humans and model organisms.
• This incompleteness hinders accurate identification of disease-related genes.

Purpose of the Study:

• To develop a novel computational method, SmuDGE, for improved disease gene prioritization.
• To generate vector-based phenotype representations using feature learning.
• To extend the applicability of phenotype-based prioritization to genes indirectly linked through interaction networks.

Main Methods:

• SmuDGE employs feature learning to create vector representations of phenotypes.
• It functions as a trainable semantic similarity measure for comparing phenotype sets (e.g., disease-gene, disease-patient).
• SmuDGE integrates interaction network data to infer phenotype representations for indirectly associated entities.

Main Results:

• SmuDGE achieves comparable or superior performance to existing semantic similarity measures in phenotype-based disease gene prioritization.
• The method significantly broadens the scope of phenotype-based approaches by including genes within connected interaction networks.
• SmuDGE effectively addresses the challenge of incomplete gene-phenotype association data.

Conclusions:

• SmuDGE offers a powerful new tool for computational disease gene prioritization.
• By utilizing interaction networks, SmuDGE overcomes data sparsity issues inherent in traditional methods.
• This approach advances the field by enabling more comprehensive gene discovery based on phenotypic information.