Causal Discovery and Optimal Experimental Design for Genome-Scale Biological Network Recovery

Area of Science:

• Genomics
• Systems Biology
• Computational Biology

Background:

• Causal discovery of genome-scale networks is crucial for understanding gene-trait relationships, including disease and drug resistance.
• Existing graphical models require interventional data and struggle to scale to the 10^3-10^4 gene range typical of genomes.

Approach:

• Introduced SP-GIES, a novel causal learner that integrates both interventional and observational datasets.
• SP-GIES demonstrates significant speedup (nearly 4x) compared to existing methods on 1,000-node networks.

Key Points:

• SP-GIES achieves a high AUC-PR score of 0.91 on 1,000-node networks.
• The method scales effectively to 2,000-node networks, representing a 4x increase in scale over previous work.
• SP-GIES enhances optimal experimental design for selecting informative interventions.

Conclusions:

• SP-GIES represents a significant advancement in scalable causal discovery for genomics.
• This work facilitates autonomous experimental design for unraveling complex biological systems.