Fence Methods

Invisible Fence (IF) for Gene Set Analysis (GSA)

We extend the fence method to situations where a true model may not exist or be among the candidate models.  This leads us to a version of the fence called the invisible fence (IF). The gene set analysis (GSA) problem is one which falls under this umbrella. Here one is interested in looking for differential expression patterns of groups or sets of genes often grouped by biological pathways. A unique fast algorithm is developed that scales efficiently and a new limited bootstrap technique derived for carrying out the analysis.

Jiang, J., Nguyen, T. and Rao, J.S. (2011).  Invisible fence methods and the identification of differentially expressed gene sets. Statistics and Its Interface, 4: 403-415.

To run a  Gene Set Analysis (GSA), first execute all of the code in the methodIF.R and IF.GSA.R functions. Then run the GSA analysis, use your data with all arguments in the function IF.GSA.

Restricted Fence (RF) for Longitudinal Data Analysis

The original fence method can be combined with the idea of restricted maximum likelihood (REML) estimation for model selection with longitudinal data.  A wild bootstrap is developed to adaptively choose the tuning parameter.

Nguyen, T. and Jiang, J. (2011). Restricted fence method for covariate selection in longitudinal data analysis. Biostatistics, 13: 303-314

To run a Restricted Fence (RF) analysis, you need to execute the codes in methodRF.R only and then use your data with all arguments in methodRF.R (i.e. there is no accompanying function for the RF analysis).

Model Selection with Missing Data

We develop a procedure for model selection in the presence of missing or incomplete data. The E-MS algorithm is in the spirit of the Expectation-Maximization (E-M) algorithm, and treats bot the model and the parameters under the model as quantities for estimation, and to be part of the iterative algorithm. Our procedure is developed under the case that the class of candidate models is finite and includes the E-MS with adaptive fence, and the E-MS with the generalized information criterion (GIC).

Jiang, J., Nguyen, T. and Rao, J.S. (2015). The E-MS algorithm: model selection with incomplete data. Journal of the American Statistical Association (in press)

Jiang, J., Nguyen, T. and Rao, J.S. (2015). Appendix to the E-MS algorithm: model selection with incomplete data. Journal of the American Statistical Association (in press)