Graph Embedding Based Novel Gene Discovery Associated With Diabetes Mellitus

Area of Science:

• Computational biology
• Genetics
• Metabolic disorders

Background:

• Diabetes mellitus is a complex metabolic disorder affecting millions globally, with unclear pathogenesis hindering effective therapies.
• Discovering novel diabetes-associated genes is crucial for a comprehensive understanding of its underlying mechanisms.
• Existing network-based methods often rely on local network structures and handcrafted features.

Purpose of the Study:

• To propose a novel computational framework for investigating novel genes associated with diabetes mellitus.
• To leverage advanced graph embedding techniques for automatic global feature extraction from molecular networks.
• To improve disease-gene prediction accuracy for diabetes mellitus.

Main Methods:

• Network feature extraction using graph embedding techniques.
• Feature denoising and regeneration via stacked autoencoders.
• Disease-gene prediction using machine learning classifiers.
• Performance comparison of different graph embedding and machine learning methods.

Main Results:

• Development of an optimized computational workflow for predicting diabetes-associated genes.
• Identification of novel candidate genes implicated in diabetes mellitus.
• Functional enrichment analysis (HPO, KEGG, GO) and publication search validated predicted genes.

Conclusions:

• The proposed framework effectively identifies novel genes associated with diabetes mellitus.
• This computational approach advances the understanding of diabetes pathogenesis.
• The findings contribute to the discovery of potential therapeutic targets for diabetes.