Scalable Bayesian variable selection for structured high-dimensional data.

Area of Science:

• Statistics
• Genomics
• Computational Biology

Background:

• Variable selection is crucial in high-dimensional data, especially in genomics.
• Existing methods struggle with scalability in large-scale genomic studies with known pathway structures.

Purpose of the Study:

• To develop a scalable Bayesian shrinkage method for variable selection in high-dimensional settings.
• To incorporate prior network information to improve model performance.

Main Methods:

• An adaptive Bayesian shrinkage approach is proposed, smoothing shrinkage parameters for connected variables.
• The model is fitted using an efficient expectation-maximization algorithm.
• Theoretical properties are established for fixed and increasing dimensions.

Main Results:

• The proposed method demonstrates advantages in variable selection and prediction.
• The approach is computationally scalable to high-dimensional settings.
• Simulations and a cancer genomics study validate the method's efficacy.

Conclusions:

• The adaptive Bayesian shrinkage method offers a scalable and effective solution for variable selection in high-dimensional genomics.
• Incorporating network information enhances model performance and interpretability.