Learning directed acyclic graphs from large-scale genomics data

Area of Science:

• Computational Biology
• Systems Biology
• Genetics

Background:

• Understanding gene interactions is crucial for deciphering complex biological processes.
• Existing methods struggle with noisy genetic data and accurately reconstructing gene networks.
• Double-knockout (DK) data offers insights but requires sophisticated analysis.

Purpose of the Study:

• To develop a robust computational method for learning genetic interaction maps from noisy DK data.
• To accurately detect and classify gene interactions and infer the underlying directed acyclic graph (DAG) topology.
• To enhance detection performance by integrating genetic interaction profile (GI-profile) data.

Main Methods:

• Proposed a linear integer optimization program, Genetic-Interactions-Detector (GENIE), for identifying gene dependencies and DAG topology.
• Extended GENIE by incorporating GI-profile data (GI-GENIE) to improve detection accuracy.
• Developed a sequential scalability technique for analyzing large gene sets and ensuring statistically significant results.

Main Results:

• GENIE accurately identifies complex biological dependencies and computes optimal DAG topologies matching DK measurements.
• GI-GENIE demonstrated superior performance in detecting gene interactions compared to standard methods.
• The sequential scalability technique provided statistically significant results for real-world gene expression data.

Conclusions:

• GENIE and GI-GENIE significantly outperform conventional techniques for genetic interaction mapping.
• The proposed methods offer accurate and scalable solutions for reconstructing gene regulatory networks from noisy biological data.
• The sequential scalability technique enables robust analysis of large-scale genetic interaction datasets.