Screening methods for linear errors-in-variables models in high dimensions.

Area of Science:

• Genomics
• Biostatistics
• Computational Biology

Background:

• Microarray studies generate high-dimensional, noisy gene expression data from limited subjects.
• Linear errors-in-variables (EIV) models are common for analyzing such data but are computationally intensive.
• Efficient variable reduction is crucial for building accurate models in high-dimensional settings.

Purpose of the Study:

• To develop computationally efficient screening procedures for high-dimensional data analyzed with linear EIV models.
• To reduce the number of features for final model building, improving scalability and performance.
• To enhance the accuracy of gene association studies.

Main Methods:

• Introduced two novel screening procedures: corrected penalized marginal screening (PMSc) and corrected sure independence screening (SISc).
• Both methods utilize corrected marginal regression models for efficient feature screening.
• The procedures are designed to handle a large number of features relative to sample size.

Main Results:

• Demonstrated that PMSc and SISc achieve screening consistency under mild conditions.
• Showed substantial feature reduction even when the number of covariates grows exponentially with sample size.
• PMSc demonstrated full variable selection consistency under weak correlation among true covariates.
• Validated the methods through simulations and analysis of gene expression data for bone mineral density.

Conclusions:

• The proposed PMSc and SISc procedures make linear EIV model estimation computationally scalable for high-dimensional data.
• These screening methods improve both estimation and selection performance in finite samples.
• The findings are applicable to gene expression analysis and other high-dimensional statistical modeling challenges.