Loading...

Controlling Group False Discovery Rate in Genome Wide Association Studies via Group Knockoffs

Asadi, Mohammad Jamal | 2025

218 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58596 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Mottahari, Abolfazl
  7. Abstract:
  8. In recent years, with the remarkable advancement of genetic technologies, a vast amount of genomic data has become available to researchers. The analysis of these data introduces various challenges, one of the most important of which is Genome-Wide Association Studies (GWAS), aiming to identify genetic loci influencing traits or diseases. Conventional methods in this field are often limited to univariate analyses or fail to account for the complex structure of genetic data, and in many cases, they do not guarantee control over the Type I error rate. In this thesis, a novel method based on the Knockoff statistical framework and its extensions is proposed. Inspired by the idea of KnockoffGAN, the introduced approach constructs group knockoff variables, providing a solution for the feature selection problem in GWAS. This method not only effectively controls the Type I error rate but also achieves higher statistical power compared to previous approaches. The obtained results show that the proposed method effectively controls the Type I error rate under most conditions and outperforms other methods in terms of statistical power. Furthermore, in the analysis of real yeast cell cycle data, the superior performance of this method was confirmed. In this experiment, various feature selection methods were evaluated at a target Type I error rate of0.3. The Group Knockoff GAN method successfully selected nine features, seven of which overlapped with known transcription factors associated with the cell cycle, whereas the Group Knockoff and Model-X Knockoff methods identified no significant features. Although the Knockoff GAN method selected eleven features, it failed to maintain proper Type I error control. These results demonstrate that incorporating group structure into the Knockoff GAN framework not only improves Type I error control but also enhances the ability to identify key factors associated with the cell cycle
  9. Keywords:
  10. Knockoff Framework ; Genome-Wide Asocciation Studies (GWAS) ; Generative Adversarial Networks ; Statistical Power ; Group False Discovery Rate

 Digital Object List

 Bookmark

...see more