GOAT: Gene-level biomarker discovery from multi-Omics data using graph ATtention neural network for eosinophilic...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Asthma is a complex, heterogeneous disease with poorly understood molecular subtypes.
• Discovering molecular biomarkers for asthma subtypes is crucial for targeted therapies.
• Multi-omics data offers potential but presents challenges due to complex inter-omics layer interactions.

Purpose of the Study:

• To develop a deep learning model for identifying molecular biomarkers of eosinophilic asthma subtypes using multi-omics data.
• To leverage graph attention neural networks to model gene interactions for biomarker discovery.

Main Methods:

• Proposed Gene-level biomarker discovery from multi-Omics data using graph ATtention neural network (GOAT), a deep attention model.
• Utilized multi-omics profiles from the COREA asthma cohort (300 patients).
• Employed graph neural networks and attention mechanisms to identify discriminating genes and model inter-gene relationships.

Main Results:

• GOAT outperformed existing models in identifying genes that discriminate eosinophilic asthma subtypes.
• The model revealed interpretable biological mechanisms underlying asthma subtypes.
• GOAT identified key genes, including transcription factors CTNNB1 and JUN, crucial for eosinophilic asthma pathophysiology, even when not distinct at the gene expression level.

Conclusions:

• GOAT is an effective deep learning approach for multi-omics biomarker discovery in complex diseases like asthma.
• The model successfully identified novel biomarkers and elucidated underlying biological mechanisms.
• Investigated transcription factors CTNNB1 and JUN highlight the model's ability to uncover subtle yet significant biological insights.