Resampling procedures to identify important SNPs using a consensus approach.

Area of Science:

• Genetics and Bioinformatics
• Statistical Genomics
• Computational Biology

Background:

• Identifying genetic variants that contribute to disease prediction is crucial for personalized medicine.
• Clinical variables alone may not capture the full spectrum of predictive information for complex traits.
• Common single-nucleotide polymorphisms (SNPs) with minor allele frequency > 1% are potential candidates for predictive modeling.

Purpose of the Study:

• To identify common SNPs that enhance predictive accuracy beyond established clinical variables.
• To develop and validate an algorithmic approach for predicting phenotypic variables using combined clinical and genotypic data.
• To assess the performance of the proposed method across different simulated phenotypic outcomes.

Main Methods:

• An algorithmic approach combining phenotypic and genotypic predictors was employed.
• Random forests were used for initial feature selection, identifying approximately 100 important SNPs.
• Least Absolute Shrinkage and Selection Operator (LASSO) with cross-validation was applied for further variable selection and shrinkage parameter tuning.

Main Results:

• The proposed procedure demonstrated good performance in predicting the Q1 outcome.
• Predictive accuracy for other outcomes was less successful, indicating outcome-specific performance.
• Resampling procedures were utilized to mitigate false positives and enhance generalizability.

Conclusions:

• The developed method effectively identifies SNPs that add predictive value, particularly for certain phenotypes.
• The combination of random forests and LASSO provides a robust framework for SNP selection in predictive modeling.
• Further investigation is needed to understand the limitations and improve performance for less predictable outcomes, especially concerning correlated false positives.