Related Experiment Video
Updated: Apr 30, 2026
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
7.0K
Stability of bivariate GWAS biomarker detection
Justin Bedő1, David Rawlinson2, Benjamin Goudey1
1NICTA Victoria Research Laboratory, University of Melbourne, Victoria, Australia; Department of Computing and Information Systems, University of Melbourne, Victoria, Australia.
Plos One
|May 3, 2014
Summary
This study introduces a new method for Genome-Wide Association Studies (GWAS) to improve the identification of true positive genetic associations. The GSS statistic demonstrated higher stability in cross-validation, outperforming traditional methods confounded by main effects.
Area of Science:
- Genetics
- Bioinformatics
- Statistical Genomics
Background:
- Genome-Wide Association Studies (GWAS) face challenges in filtering false positives among candidate causal SNPs.
- Traditional methods often struggle with the computational demands of exhaustive searches for complex genetic interactions.
- Cross-validation is crucial for assessing the reliability and stability of feature selection methods in GWAS.
Purpose of the Study:
- To evaluate the stability of bivariate feature selection methods in GWAS using cross-validation.
- To compare the performance of the GSS statistic against traditional chi-squared tests and GBOOST.
- To investigate the impact of univariately significant SNPs on the replicability of bivariate association tests.
Main Methods:
- Performed 10 trials of 2-fold cross-validation on seven GWAS datasets.
- Compared the chi-squared test, GBOOST, and the GSS statistic for exhaustive bivariate analysis.
- Utilized Spearman's correlation and an extended version to measure rank list similarity.
Main Results:
- The GSS statistic showed higher stability across cross-validation folds for most diseases compared to the chi-squared test.
- The traditional chi-squared test was significantly confounded by univariately significant SNPs (main effects).
- GSS and GBOOST stability remained unaffected by the removal of univariately significant SNPs, indicating a focus on true bivariate associations.
Conclusions:
- The GSS statistic offers a more stable and reliable approach for identifying bivariate SNP-pair associations in GWAS.
- GSS and GBOOST successfully target true bivariate associations, unlike the confounded chi-squared test.
- GSS demonstrates potential for detecting a larger set of relevant SNP-pairs compared to GBOOST.
