Variable Selection for Sparse High-Dimensional Nonlinear Regression Models by Combining Nonnegative Garrote and Sure...

Area of Science:

• Statistics
• Bioinformatics
• Computational Biology

Background:

• Nonlinear relationships between covariates and responses are common in regression.
• Reconstructing gene regulatory networks requires identifying key genetic factors.

Purpose of the Study:

• To develop a method for simultaneous variable selection and parameter estimation in sparse, high-dimensional nonlinear additive regression models.
• To address the challenge of identifying important covariates in gene regulatory network reconstruction.

Main Methods:

• Proposed a novel iterative approach combining nonlinear independence screening (NLIS) for large-scale variable screening and nonnegative garrote (NNG) for moderate-scale model selection.
• NLIS procedure demonstrates a sure screening property and handles non-polynomial dimensionality.
• NNG for nonlinear additive regressions ensures selection consistency for unimportant covariates and estimation consistency for important ones.

Main Results:

• The NLIS procedure effectively screens relevant variables, even with complex, non-polynomial relationships.
• The NNG method accurately selects important covariates and estimates their parameters consistently.
• The combined method shows strong numerical performance on simulated and real gene expression data.

Conclusions:

• The proposed method offers a robust solution for variable selection and parameter estimation in nonlinear additive regression.
• This approach is particularly valuable for applications like gene regulatory network analysis, improving the identification of causal genetic interactions.