Reflection Knockoffs via Householder Reflection: Applications in Proteomics and Genetic Fine Mapping.

Area of Science:

• Statistics
• Genomics
• Proteomics

Background:

• High-dimensional datasets often contain highly correlated features, complicating accurate statistical inference.
• Existing knockoff methods struggle with feature selection when feature correlation is substantial.

Purpose of the Study:

• Introduce a novel knockoff construction method, reflection knockoffs, for superior feature selection.
• Enhance the consistency of knockoff filters through aggregation of knockoff statistics.
• Apply the new method to identify proteomic signatures of age and perform genetic fine mapping.

Main Methods:

• Construct reflection knockoffs using Householder reflections of original features.
• Aggregate multiple sets of identically distributed knockoff statistics.
• Utilize knockoff filters with reflection knockoffs for feature selection in biological applications.

Main Results:

• Reflection knockoffs significantly outperform Model-X knockoffs in feature selection accuracy, particularly in highly correlated settings.
• Analysis of proteomic data revealed many initially identified age associations were non-drivers (hitchhikers).
• Knockoff filters with reflection knockoffs and aggregation surpassed state-of-the-art methods in genetic fine mapping.

Conclusions:

• Reflection knockoffs offer a powerful and more accurate approach to feature selection in high-dimensional, correlated data.
• The method has significant implications for distinguishing true biological drivers from spurious associations.
• Potential applications include privacy-preserving genetic data sharing.