Gene regulatory network inference using mixed-norms regularized multivariate model with covariance selection

Area of Science:

• Systems Biology
• Computational Biology
• Bioinformatics

Background:

• Reconstructing gene regulatory networks (GRNs) from transcriptomics data is a significant challenge.
• Non-linear methods improve GRN reconstruction, but the benefit of jointly modeling multiple target genes under linear assumptions is unclear.

Purpose of the Study:

• To develop and evaluate novel methods for GRN reconstruction by jointly modeling multiple target genes.
• To assess if simultaneous modeling enhances GRN inference accuracy compared to existing approaches.

Main Methods:

• Proposed two novel approaches blending regularized multivariate regression and graphical models.
• Utilized L2,1-norm with classical regularization techniques for GRN reconstruction.
• Validated models using DREAM5 challenge data and datasets from Escherichia coli and Saccharomyces cerevisiae.

Main Results:

• The proposed models demonstrated consistently good performance across diverse datasets.
• Achieved improved GRN inference accuracy through simultaneous modeling of multiple target genes.
• Successfully identified master regulators consistent with experimental evidence in E. coli.

Conclusions:

• Simultaneous modeling of multiple target genes significantly improves GRN inference.
• The developed methods offer a reliable alternative for GRN reconstruction in systems biology.
• The approach facilitates accurate prediction and analysis of master regulator plasticity.