Ultra high-dimensional semiparametric longitudinal data analysis

Area of Science:

• Statistics
• Bioinformatics
• Public Health

Background:

• Ultra-high-dimensional longitudinal data present challenges in fields like public health and bioinformatics.
• Existing methods struggle with models where covariate dimension grows exponentially with sample size.
• Flexible and sparse modeling approaches are needed for these complex datasets.

Purpose of the Study:

• To develop a flexible semiparametric approach for ultra-high-dimensional longitudinal data.
• To address challenges posed by exponentially growing covariate dimensions.
• To enable simultaneous variable selection and estimation in complex data settings.

Main Methods:

• Partially linear single-index models are employed for ultra-high-dimensional longitudinal data.
• Penalized generalized estimating equations (GEE) are utilized.
• A smoothly clipped absolute deviation (SCAD) penalty is applied for variable selection and estimation.

Main Results:

• The proposed method effectively handles ultra-high-dimensional partially linear and single-index covariates.
• Simultaneous variable selection and estimation are achieved with the SCAD penalty.
• Asymptotic theory, including the oracle property, is established for both components in ultra-high dimensions.
• An efficient algorithm is presented to address computational challenges.

Conclusions:

• The developed partially linear single-index model offers a powerful tool for analyzing ultra-high-dimensional longitudinal data.
• The method demonstrates effectiveness in capturing correlations, nonlinearity, and interactions.
• Validation through simulation studies and a yeast cell cycle gene expression dataset confirms the approach's utility.