Speos: an ensemble graph representation learning framework to predict core gene candidates for complex diseases

Area of Science:

• Genomics
• Systems Biology
• Computational Biology

Background:

• Understanding phenotype-genotype relationships is key in biology and medicine.
• The omnigenic model suggests core genes mediate trait effects, influenced by peripheral regulatory networks.
• Identifying core genes is essential for understanding disease mechanisms.

Purpose of the Study:

• To develop a computational method for predicting core-like genes associated with diseases.
• To validate the predicted core-like genes using external biological data.
• To identify novel druggable targets for therapeutic development.

Main Methods:

• A positive-unlabeled graph representation learning ensemble approach was developed.
• Nested cross-validation was used for training and prediction.
• Mouse knockout phenotypes were employed for external validation.

Main Results:

• Predicted core-like genes showed disease-relevant phenotypes in mouse knockouts, similar to Mendelian genes.
• Candidate genes exhibited core gene properties: transcriptional deregulation and loss-of-function intolerance.
• Identified genes are enriched for druggable targets, including many currently untargeted ones.

Conclusions:

• The graph learning approach effectively predicts core-like genes.
• These predicted genes represent promising, druggable targets for future drug development.
• The study highlights the potential of computational methods for biological interpretation and drug discovery.