Genome Wide Epistasis Study of On-Statin Cardiovascular Events with Iterative Feature Reduction and Selection

Area of Science:

• Cardiovascular Genetics
• Pharmacogenomics
• Machine Learning in Medicine

Background:

• Predicting major adverse cardiovascular events (MACE) is crucial, but risk stratification after statin initiation is challenging.
• Current genetic risk factors for on-statin MACE have limited effect size and generalizability.
• High-level epistatic interactions (gene-gene interactions) may offer better predictive power for on-statin MACE.

Purpose of the Study:

• To identify high-level epistatic risk factors for on-statin MACE using genome-wide association study (GWAS)-scale data.
• To develop an interpretable method for risk prediction in patients on statin therapy.
• To explore the utility of machine learning for uncovering complex genetic risk patterns.

Main Methods:

• Utilized controlled-access data from 5890 statin users from Vanderbilt University Medical Center's BioVU.
• Employed Random Forest Iterative Feature Reduction and Selection (RF-IFRS) on GWAS-scale data to identify informative genetic and environmental features.
• Constructed gene-variant networks and decision trees from selected variant pairs for risk interpretation.

Main Results:

• Identified six gene-variant networks associated with predicting the odds of on-statin MACE.
• Pathway analysis revealed associations between variants in vasculogenesis, angiogenesis, and carotid artery disease genes and on-statin MACE risk.
• A subset of patients with variants in *COL4A2*, *TMEM178B*, *SZT2*, and *TBXAS1* showed significantly elevated odds of on-statin MACE (OR = 4.53).

Conclusions:

• The RF-IFRS method effectively interprets complex machine learning outputs for identifying epistatic networks.
• These identified epistatic networks show potential for improving risk estimation for on-statin MACE.
• Further research is needed to validate these findings in diverse populations.