Identifying Network Perturbation in Cancer.

Area of Science:

• Computational biology
• Systems biology
• Genomics

Background:

• Gene regulatory networks (GRNs) are crucial for understanding biological processes.
• Identifying dynamic changes in GRNs between distinct biological states, such as disease versus normal, is challenging.

Purpose of the Study:

• To develop a computational framework, DISCERN (DIfferential SparsE Regulatory Network), for identifying topological changes in GRNs.
• To accurately detect perturbed genes and uncover underlying mechanisms in disease states using gene expression data.

Main Methods:

• DISCERN infers conditional dependencies between candidate regulators and genes, improving accuracy over pairwise correlation.
• A likelihood-based scoring function is employed to enhance the accuracy of inferred network edges.
• The framework compares gene expression datasets from diseased and normal tissues to identify differential network structures.

Main Results:

• DISCERN demonstrated higher accuracy in identifying perturbed genes on synthetic data compared to existing methods.
• In acute myeloid leukemia, breast cancer, and lung cancer datasets, high-scoring genes were enriched for known tumor drivers and disease-associated processes.
• DISCERN explained observed epigenomic activity patterns more accurately than alternative methods.

Conclusions:

• DISCERN provides a robust computational approach for identifying differential gene regulatory networks.
• The framework accurately identifies disease-perturbed genes and offers insights into disease mechanisms.
• DISCERN has potential applications in understanding cancer biology and patient prognosis.