Genetic risk assessment based on association and prediction studies

Area of Science:

• Genetics
• Biomarker Discovery
• Personalized Medicine

Background:

• Genetic factors influence disease risk and progression.
• Genome-Wide Association Studies (GWAS) identify population-level genetic associations.
• Individual-level risk prediction requires complementary approaches like machine learning (ML).

Purpose of the Study:

• To develop and evaluate personalized risk assessment models for predicting hepatitis B surface antigen (HBsAg) seroclearance.
• To compare the predictive power of GWAS-identified and ML-identified biomarkers.
• To assess the synergistic effect of combining GWAS and ML biomarkers.

Main Methods:

• Utilized single-nucleotide polymorphism (SNP) data from 200 Korean patients (100 HBsAg seroclearance, 100 high HBsAg).
• Developed risk models using GWAS-identified candidate biomarkers and ML (Random Forest)-identified biomarkers.
• Evaluated model accuracy with individual and combined biomarker sets.

Main Results:

• Models using relevant biomarkers showed higher accuracy than using all features (64%).
• GWAS biomarkers achieved 82% (52 markers) and 71% (3 markers) accuracy.
• ML biomarkers achieved 80% accuracy (150 markers).
• Combining GWAS and ML biomarkers improved predictive accuracy to 90%.

Conclusions:

• Relevant biomarkers significantly influence phenotypic emergence.
• ML serves as a valuable auxiliary analysis to GWAS for enhanced predictive modeling.
• ML-identified biomarkers are linked to hepatocellular carcinoma (HCC), even in HBsAg-negative cases.