glmgraph

glmgraph implements network-constrained sparse regression to perform variable selection and predictive modeling of high-dimensional genomic and omics data.

Key Features:

• Network-Constrained Sparse Regression: Leverages graph or network structures in omics data to incorporate prior biological knowledge into variable selection and modeling.
• Sparse Linear and Logistic Regression: Supports both sparse linear regression and sparse logistic regression models for continuous and binary outcomes.
• Regularization Techniques: Provides L1 penalty (Lasso) for sparsity, Minimax Concave Penalty (MCP) to reduce bias, and Laplacian penalty for coefficient smoothing across connected variables.
• Efficient Optimization Algorithm: Solves the penalized regression problems using a coordinate descent algorithm.
• Implementation: Implemented in R with performance-critical code in C++ using the Armadillo library.

Scientific Applications:

• Personalized Medicine: Relates high-dimensional omics features to phenotypes for predictive modeling and biomarker selection.
• Microbiome Studies: Utilizes phylogenetic relationships among bacterial taxa to improve identification of relevant features and prediction.
• Network-based Genomics: Integrates gene regulatory networks and protein-protein interaction networks into regression analyses to enhance interpretability and performance.

Methodology:

Integrates graph/network constraints into sparse regression models; supports sparse linear and logistic regression with L1 (Lasso), Minimax Concave Penalty (MCP), and Laplacian penalties; optimization via coordinate descent; implemented in R with C++ using the Armadillo library.