Gene Self-Expressive Networks as a Generalization-Aware Tool to Model Gene Regulatory Networks

Area of Science:

• Computational Biology
• Network Inference
• Systems Biology

Background:

• Self-expressiveness is a mathematical property for characterizing data relationships, successfully applied in various fields.
• Its potential for inferring gene networks remains largely unexplored.
• Existing methods for gene network inference may lack interpretability or generalization capabilities.

Purpose of the Study:

• To introduce Generalizable Gene Self-Expressive Networks (GXN) as a novel formalism for gene network modeling.
• To propose two new methods, GXN•EN and GXN•OMP, for inferring and assessing these networks.
• To evaluate the performance and interpretability of GXN methods on diverse biological datasets.

Main Methods:

• Developed GXN, a generalization-aware formalism for gene network inference.
• Implemented GXN•EN using ElasticNet and GXN•OMP using Orthogonal Matching Pursuit (OMP).
• Evaluated methods on Microarray (DREAM5) and RNA-seq datasets, including a comparison of normal vs. Alzheimer's disease conditions.

Main Results:

• GXN•EN and GXN•OMP achieved performance comparable to state-of-the-art tools.
• Methods demonstrated fast training times and high sparsity, enhancing interpretability.
• Successfully detected differential gene sub-networks between normal and disease states, revealing biologically relevant communities.

Conclusions:

• GXN provides an interpretable and generalization-aware approach to gene network inference.
• The proposed methods are efficient and yield sparse, modular gene networks.
• Inferred networks highlight functional gene communities and offer insights into disease-specific alterations.