Recursive random forest algorithm for constructing multilayered hierarchical gene regulatory networks that govern...

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Multilayered hierarchical gene regulatory networks (ML-hGRNs) are crucial for understanding biological pathway regulation.
• Existing computational methods for constructing ML-hGRNs are limited.

Purpose of the Study:

• To develop a novel computational algorithm for the direct construction of ML-hGRNs.
• To improve the accuracy and efficiency of ML-hGRN inference.

Main Methods:

• Developed a Backward Elimination Random Forest (BWERF) algorithm.
• Utilized random forest models and backward elimination to identify transcription factor (TF) importance.
• Employed Gaussian mixture models for TF retention and layer-wise network construction.

Main Results:

• The BWERF algorithm successfully constructed ML-hGRNs for mouse pluripotency and Arabidopsis pathways.
• BWERF demonstrated improved accuracy in identifying authentic TFs compared to GENIE3.
• BWERF generated ML-hGRNs with fewer edges, facilitating experimental validation.

Conclusions:

• BWERF enhances the accuracy of ML-hGRN construction by reducing noise and effectively aggregating TF importance.
• The algorithm provides a more refined network structure, aiding biologists in selecting regulatory edges for validation.
• BWERF represents a significant advancement in computational approaches for dissecting complex gene regulatory architectures.