Large-scale, interpretable gene regulatory network inference through biologically informed matrix factorization

Area of Science:

• Computational Biology
• Systems Biology
• Genomics

Background:

• Gene regulatory networks (GRNs) are crucial for understanding cellular identity and phenotype.
• Existing methods for GRN inference primarily focus on identifying regulatory interactions.
• A complementary approach is needed to determine the direction (activating or inhibitory) of these regulatory effects.

Purpose of the Study:

• To develop a novel computational framework, Giraffe, for inferring gene regulatory networks (GRNs).
• To jointly estimate transcription factor (TF) activities and the directionality of regulatory effects.
• To provide a scalable and computationally efficient method for mechanistic interpretation of GRNs.

Main Methods:

• Developed Giraffe, a matrix factorization framework integrating gene expression, motif-based regulatory priors, and TF-protein interactions.
• Giraffe estimates signed partial regulatory effects, quantifying the strength and direction of TF influence.
• The framework builds upon existing methods like PANDA, offering enhanced mechanistic insights.

Main Results:

• Giraffe accurately reconstructs regulatory interactions across synthetic data, human tissues, yeast perturbation experiments, and cancer.
• The method successfully distinguishes between activating and inhibitory regulatory effects with high accuracy.
• Inferred networks captured tissue-specific regulation, correctly classified TF perturbation effects, and identified cancer-related regulatory changes.

Conclusions:

• Estimating the direction of transcriptional regulation offers a valuable complementary perspective to traditional GRN inference.
• Giraffe provides a scalable, flexible, and computationally efficient tool for detailed GRN analysis.
• The framework facilitates biological interpretation and hypothesis generation for complex regulatory systems.