Globally learning gene regulatory networks based on hidden atomic regulators from transcriptomic big data.

Area of Science:

• Systems Biology
• Bioinformatics
• Computational Biology

Background:

• Gene regulation involves numerous unknown or unobserved regulators.
• Existing gene regulatory network (GRN) inference methods often provide local and sub-optimal results, neglecting unknown regulators.

Purpose of the Study:

• To propose a global GRN inference framework, dlGRN, utilizing dictionary learning.
• To address limitations of current methods by inferring regulatory relationships in a global manner.

Main Methods:

• Developed a modified dictionary learning (DL) algorithm to learn atomic regulators (ARs) from gene expression data.
• Implemented a global regression approach to predict regulatory relationships.
• Ensured the DL algorithm aligns with the scale-free property of biological networks.

Main Results:

• dlGRN effectively infers GRNs by learning ARs and predicting regulations globally.
• The method intrinsically distinguishes between direct and indirect regulatory interactions.
• Experimental validation confirmed a novel transcription factor TFAP2C and oncogene EGFR regulation in lung cancer cells.

Conclusions:

• dlGRN demonstrates effectiveness and efficiency in GRN reverse engineering on both simulated and real-world data.
• The study highlights the significance of DNA methylation in gene regulatory systems.
• dlGRN is a robust and global tool suitable for standalone GRN inference.