glmgraph: an R package for variable selection and predictive modeling of structured genomic data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput genomic data analysis aims to identify relevant features and build predictive models for personalized medicine.
• The small sample size (n) and high dimensionality (p) in omics data present a significant challenge for correlating features with phenotypes.
• Sparse regression models, particularly network-constrained ones, leverage prior biological network structures to address the small n, large p problem.

Purpose of the Study:

• To develop an efficient R software package for graph-constrained regularization in sparse regression.
• To provide an implementation for both sparse linear and logistic regression models.
• To facilitate the analysis of omics data by incorporating network structures.

Main Methods:

• The 'glmgraph' package implements graph-constrained regularization using L1 penalty and minimax concave penalty for variable selection.
• Laplacian penalty is utilized for coefficient smoothing.
• An efficient coordinate descent algorithm is employed to solve the optimization problems.
• The package is implemented in R and C++ Armadillo and available on CRAN.

Main Results:

• The 'glmgraph' package offers an efficient R implementation for network-constrained sparse regression.
• It supports both sparse linear and logistic regression with various penalty options.
• Demonstrated application on a human microbiome dataset highlights its utility in analyzing data with known phylogenetic structures.

Conclusions:

• 'glmgraph' provides a valuable tool for researchers analyzing high-dimensional omics data with network information.
• The package enhances feature selection and predictive modeling capabilities in genomics.
• Its availability on CRAN promotes wider adoption and application in personalized medicine and other omics-driven fields.