Expanding drug targets for 112 chronic diseases using a machine learning-assisted genetic priority score

Area of Science:

• Genetics
• Pharmacology
• Computational Biology

Background:

• Identifying genetic drivers of chronic diseases is crucial for advancing drug discovery.
• Existing methods for genetic target discovery can be limited in scope and predictive power.

Purpose of the Study:

• To develop and validate a machine learning-assisted genetic priority score (ML-GPS) for enhanced chronic disease target discovery.
• To integrate genetic associations with predicted disease phenotypes to improve the identification of novel drug targets.

Main Methods:

• Constructed gradient boosting models to predict 112 chronic disease phecodes in the UK Biobank.
• Analyzed associations between predicted/observed phenotypes and genetic variants across the allele frequency spectrum.
• Integrated genetic associations with existing evidence using gradient boosting to create ML-GPS, trained on drug indications from Open Targets and tested on SIDER data.

Main Results:

• Generated ML-GPS predictions for over 2.3 million gene-phecode pairs.
• Predicted phenotypes identified substantially more genetic associations than observed phenotypes, significantly improving ML-GPS performance.
• The top 1% of ML-GPS scores supported 15,077 previously unsupported gene-phecode pairs, increasing drug target coverage.

Conclusions:

• ML-GPS effectively enhances the discovery of genetic drivers for chronic diseases.
• The score successfully identifies known target-disease relationships and promising novel targets, including those for drugs in clinical trials.
• ML-GPS represents a powerful tool for prioritizing genetic targets in drug discovery pipelines.