RCFGL: Rapid Condition adaptive Fused Graphical Lasso and application to modeling brain region co-expression networks

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• Gene co-expression networks are crucial for understanding gene regulation and pathway activity.
• Jointly estimating networks across multiple conditions enhances analytical power and reveals condition-specific patterns.
• Existing methods like Condition Adaptive Fused Graphical Lasso (CFGL) face computational limitations.

Purpose of the Study:

• To propose a faster and more scalable alternative to CFGL for joint gene co-expression network estimation.
• To improve computational efficiency and generalize the method for more than three conditions.
• To introduce a screening rule for reducing computational complexity by identifying disjoint sub-networks.

Main Methods:

• Developed Rapid Condition Adaptive Fused Graphical Lasso (RCFGL) by incorporating condition specificity into the Fused Multiple Graphical Lasso (FMGL) model.
• Employed a more efficient algorithm with a computational complexity of O(p2K).
• Introduced a novel screening rule to decompose the network estimation problem into smaller sub-networks.

Main Results:

• RCFGL demonstrates significant computational advantages and superior performance compared to FGL, FMGL, and CFGL in simulations and real data.
• The method successfully estimated gene co-expression networks across different rat brain regions.
• The proposed method is easily generalizable for more than three conditions.

Conclusions:

• RCFGL offers a computationally efficient and scalable solution for joint gene co-expression network inference.
• The method provides valuable insights into condition-specific gene regulation and pathway activity.
• An accessible C and Python package is available for implementing RCFGL.