Variable Selection with Prior Information for Generalized Linear Models via the Prior LASSO Method.

Area of Science:

• Statistical modeling
• Bioinformatics
• Genomics

Background:

• LASSO is a statistical method for variable selection and parameter estimation in generalized linear models.
• High-dimensional biological and biomedical data often contain valuable prior information.
• Standard LASSO may have limited power with a large number of variables.

Purpose of the Study:

• To propose an extension of LASSO, named prior LASSO (pLASSO).
• To incorporate prior information into penalized generalized linear models.
• To improve variable selection and parameter estimation in high-dimensional data.

Main Methods:

• Developed pLASSO by adding a discrepancy measure between prior information and the model to the LASSO criterion.
• Utilized a Least Angle Regression (LARS)-like procedure for the pLASSO estimator's solution path in linear regression.
• Employed asymptotic theories and simulation studies to evaluate performance.

Main Results:

• pLASSO significantly outperforms standard LASSO when prior information is accurate.
• pLASSO demonstrates robustness to misspecified or less reliable prior information.
• The method was illustrated using a genome-wide association study dataset.

Conclusions:

• pLASSO offers a powerful approach to leverage prior biological and biomedical data for enhanced variable selection.
• The method provides a robust and effective alternative to standard LASSO in high-dimensional settings.
• pLASSO has practical applications in fields like genome-wide association studies.