PAGE: an R package for network detection of multivariate error-prone gene expression data with the availability of auxiliary information

Gene expression data in bioinformatics studies often contain multivariate or high-dimensional variables. One key research problem is to uncover the network structure among gene expression variables, which can help identify pathway-level disruptions associated with diseases and support the development of targeted therapies. With the increasing availability of auxiliary variables (known as covariates), it is desirable to incorporate them to enhance network detection of the main variables (known as responses). The main challenge lies in accurately selecting informative covariates and recovering the network structure among responses, especially when using linear or nonlinear models to characterize the relationships between multivariate responses and covariates. Another challenge is the presence of measurement error in gene expression data, which may result from limitations in measurement precision or human recording errors.

To address these challenges and provide a reliable, publicly accessible analytical tool, we develop an R package named PAGE. This package includes three core functions that support measurement error correction, variable selection, and network estimation under both linear and nonlinear modeling frameworks. The application of PAGE is demonstrated using a yeast cell cycle dataset.

Based on the analysis and demonstration, we find that the R package PAGE is valid for dealing with the complex network structure. In addition, throughout the simulation studies, we find that the correction of measurement error is crucial, and the R package PAGE is useful to handle error-prone data.