New AI Method Improves Disease Risk Prediction Using Biological Markers
EndoPRS combines genetic data with biological markers to better predict childhood asthma risk, outperforming existing methods.
Summary
Researchers developed EndoPRS, a machine learning method that improves disease risk prediction by incorporating endophenotypes—biological markers linked to disease development. Testing on childhood asthma prediction using eosinophil counts, EndoPRS significantly outperformed existing polygenic risk score methods in both accuracy and transferability across populations. This approach addresses limitations of current multi-trait methods that don't account for how some traits mediate disease development.
Detailed Summary
Predicting disease risk from genetic data has become increasingly sophisticated, but current methods miss important biological relationships. Most polygenic risk scores treat all related traits equally, ignoring that some biological markers actually mediate disease development rather than just correlating with it.
Researchers at University of North Carolina developed EndoPRS, a weighted lasso regression model that incorporates endophenotypes—biological traits that lie on the causal pathway between genes and disease. Unlike existing multi-trait methods, EndoPRS doesn't assume specific genetic architectures, making it more robust across different disease mechanisms.
Testing on childhood-onset asthma prediction, the team used eosinophil count (a type of white blood cell) as an endophenotype, since elevated eosinophils are known to contribute to asthma development. They validated EndoPRS using UK Biobank and All of Us datasets, comparing it against standard polygenic risk scores and multi-trait methods like MTAG and wMT-BLUP.
EndoPRS significantly improved prediction accuracy and showed better transferability across different populations compared to existing methods. The approach was particularly robust across various genetic architectures in simulation studies, suggesting broad applicability beyond asthma.
This advancement could enhance personalized medicine by providing more accurate disease risk predictions, especially valuable for early intervention strategies. The method's ability to incorporate biological knowledge about disease pathways represents a meaningful step toward more clinically useful genetic risk assessment tools.
Key Findings
- EndoPRS outperformed existing polygenic risk score methods for childhood asthma prediction
- Method showed superior transferability across different population datasets
- Approach remained robust across various genetic architecture scenarios in simulations
- Incorporating eosinophil count as endophenotype significantly improved asthma risk prediction
- Method doesn't require assumptions about underlying genetic relationships
Methodology
Weighted lasso regression model incorporating endophenotype data with genetic variants. Validated using UK Biobank and All of Us datasets with childhood-onset asthma and eosinophil count data. Extensive simulation studies tested robustness across different genetic architectures.
Study Limitations
Study focused specifically on asthma and eosinophils; broader disease applications need validation. Method requires availability of relevant endophenotype data, which may not always be accessible in clinical settings.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.