Self Supervised Prediction of Genetic Associations in Comorbid Diseases With Masked Autoencoder Using Hypergraph...

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Comorbid diseases, the simultaneous occurrence of multiple diseases, present complex genetic traits.
• Understanding the genetic basis of comorbid diseases is crucial for unraveling their pathogenesis.
• Current methods for predicting genetic associations in comorbid diseases often lack multi-relational data integration and a unified approach.

Purpose of the Study:

• To develop a generalized and novel approach for predicting overlapping genetic associations across comorbid diseases.
• To leverage gene co-expression networks and hypergraph learning for enhanced biological data integration.
• To establish a unified method for identifying common genetic drivers of comorbid conditions.

Main Methods:

• A novel approach combining hypergraph-based pre-embedding learning with a self-supervised edge-masking technique.
• Utilizing disease-specific gene co-expression networks to infer common genetic associations.
• Employing hypergraph learning to capture higher-order biological information from candidate genes.
• Implementing self-supervised learning for model training with limited edge labels.

Main Results:

• The proposed approach successfully predicts overlapping genetic associations from gene co-expression networks.
• The method outperforms six baseline models on case-study datasets.
• Novel genetic associations across comorbid disease pairs were identified.

Conclusions:

• The developed approach provides a unified and effective method for predicting common genetic associations in comorbid diseases.
• Hypergraph learning and self-supervised edge masking enhance the prediction of genetic links by integrating rich biological information.
• This work contributes to a deeper understanding of the molecular basis of complex, co-occurring diseases.