Identifying Prognostic SNPs in Clinical Cohorts: Complementing Univariate Analyses by Resampling and Multivariable...

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Clinical cohorts with time-to-event endpoints are increasingly analyzed using large numbers of single nucleotide polymorphisms (SNPs).
• Limited cohort sizes necessitate advanced analysis strategies for identifying prognostic SNPs and characterizing disease processes.
• Existing methods often struggle with the high dimensionality of SNP data relative to sample size.

Purpose of the Study:

• To propose a novel analysis strategy for stable selection of prognostic SNPs in clinical cohorts.
• To compare a multivariable regression approach with univariate testing for SNP selection.
• To identify a small, validated set of SNPs and genes for better disease characterization.

Main Methods:

• A strategy combining univariate testing (permutation-based gene-level testing) and multivariable regularized regression models.
• Simultaneous consideration of all SNPs using regularized regression for prognostic SNP selection.
• Stability assessment via resampling inclusion frequencies for both univariate and multivariable approaches.

Main Results:

• The multivariable regression approach automatically focused on smaller sets of SNPs, often corresponding to blocks of correlated SNPs.
• This targeted extraction led to more stable SNP and gene selection compared to the univariate approach.
• The strategy was successfully illustrated with acute myeloid leukemia patient data and validated through simulation.

Conclusions:

• Regularized multivariable regression with resampling offers a robust method for stable SNP selection in high-dimensional clinical cohort data.
• This approach enhances the characterization of disease processes by identifying key prognostic SNPs.
• The proposed strategy provides a valuable alternative to traditional univariate analyses for SNP data in clinical research.