High-dimensional generalized median adaptive lasso with application to omics data.

Area of Science:

• Statistics
• Biostatistics
• Genomics

Background:

• High-dimensional data analysis presents challenges in variable selection for causal inference.
• Skewed outcome distributions can compromise the accuracy of traditional causal effect estimation methods.
• Accurate covariate selection is crucial for reliable causal inference.

Purpose of the Study:

• To introduce a novel method, the generalized median adaptive lasso (GMAL), for robust variable selection in causal inference.
• To address the challenges posed by skewed outcome distributions in high-dimensional data.
• To improve the accuracy of causal effect estimation under skewed data conditions.

Main Methods:

• Developed the generalized median adaptive lasso (GMAL) for covariate selection.
• Utilized a linear median regression model for constructing penalty weights within GMAL.
• Evaluated GMAL's performance through simulations and application to a real-world dataset.

Main Results:

• GMAL demonstrated comparable variable selection performance to existing methods with symmetric outcome distributions.
• GMAL showed superior performance in variable selection when outcome distributions were skewed.
• GMAL consistently outperformed existing methods in causal effect estimation, evidenced by lower root-mean-square error.

Conclusions:

• GMAL effectively handles skewed outcome distributions, ensuring accurate variable selection and causal effect estimation.
• The proposed method offers a significant advancement for causal inference in high-dimensional settings with non-normally distributed outcomes.
• GMAL was applied to a DNA methylation dataset to explore the relationship between cerebrospinal fluid tau protein and Alzheimer's disease severity.