Variable Selection in Heterogeneous Datasets: A Truncated-rank Sparse Linear Mixed Model with Applications to...

Area of Science:

• Genomics
• Statistical Genetics
• Bioinformatics

Background:

• High-dimensional datasets, common in biology and medicine, present challenges for traditional variable selection.
• Existing methods like Lasso may produce numerous false discoveries in datasets with complex, non-independently and identically distributed (non-i.i.d.) structures.
• Variable selection is crucial for understanding genetic associations, especially when population structure is unknown.

Purpose of the Study:

• To develop a robust framework for sparse variable selection in datasets with unknown subpopulations.
• To adaptively correct for population structure without prior knowledge of individual relationships.
• To improve the accuracy of variable selection in complex genomic data.

Main Methods:

• A unified framework for sparse variable selection is proposed.
• A low-rank linear mixed model is utilized to adaptively correct for population structure.
• The method adaptively selects an appropriate covariance structure complexity.

Main Results:

• The proposed framework demonstrates superior performance compared to existing methods in extensive experiments.
• The method effectively corrects for population structure, leading to more reliable variable selection.
• Validation on plant, mouse, and human genomic datasets confirms its practical utility.

Conclusions:

• The developed framework offers a powerful solution for variable selection in structured, high-dimensional biological data.
• It enhances the discovery of true associations by mitigating the impact of hidden population structures.
• The method provides a flexible and effective tool for genomic research across different species.