netReg: network-regularized linear models for biological association studies.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Linear regression in high-dimensional genomic datasets (n ≪ p) is challenging due to the large number of variables and limited observations.
• Existing penalized regression methods address some issues, but models incorporating prior biological network knowledge for multivariate genomic responses are lacking efficient implementations.

Purpose of the Study:

• To develop and implement netReg, a computationally efficient, freely available package for graph-penalized regression.
• To integrate prior biological network information into linear models for improved regression coefficient estimation in high-dimensional genomic data.

Main Methods:

• netReg utilizes graph-penalized regression models to incorporate biological network information.
• The package is implemented in C++ for core computations, leveraging Armadillo for matrix calculations and Dlib for optimization.
• It is available as an R-package on Bioconductor and a C++ command-line tool via Bioconda.

Main Results:

• netReg enables the analysis of large networks with thousands of variables.
• The package provides sparse or smooth solutions for regression coefficients by incorporating a priori biological graph information.
• Computational efficiency is achieved through C++ implementation with optimized libraries.

Conclusions:

• netReg offers a novel and efficient solution for modeling biological associations in high-dimensional genomic data.
• The package facilitates the use of prior biological network knowledge to enhance the reliability of regression coefficient estimates.
• Freely available as an R-package and C++ tool, netReg supports broader application in genomic research.