Machine learning models incorporating genotype and ancestry improve severe asthma risk prediction.

Area of Science:

• Genomics
• Machine Learning
• Personalized Medicine

Background:

• Severe asthma (SA) poses challenges, especially in African American pediatric populations with poor response to inhaled corticosteroids (ICS).
• Predicting treatment response is crucial for effective personalized medicine.

Purpose of the Study:

• To develop and evaluate a novel machine learning (ML)-based stacking technique integrating Single Nucleotide Polymorphisms (SNPs) and local ancestry (LA) for improved prediction of ICS response in SA.
• To assess the complementary value of SNP and LA data in predicting clinical outcomes.

Main Methods:

• Utilized whole-exome sequenced data from an African American pediatric cohort (N=248) from the Biorepository and Integrative Genomics (BIG) Initiative.
• Developed ML pipelines using SNP and LA data, employing stratified 10-fold cross-validation and stacking techniques.
• Integrated logistic regression and random forest models with feature selection for predictive modeling.

Main Results:

• The integrated SNP and LA data achieved a significantly higher Area Under the Curve (AUC) of 0.729 ± 0.048 compared to SNP (0.693 ± 0.066) or LA (0.625 ± 0.103) data alone (p=0.005).
• Both SNP and LA data provided distinct and complementary insights, with different key contributing features.
• LA data alone showed comparable predictive performance to SNP data alone.

Conclusions:

• The proposed stacking ML technique effectively integrates SNP and LA data, enhancing predictive accuracy for medication response in complex conditions like SA.
• This holistic approach holds significant promise for advancing personalized medicine by leveraging multifactorial genomic and ancestry data.
• The findings underscore the importance of combining diverse data types for robust clinical outcome prediction.