Optimal adjustment sets for causal query estimation in partially observed biomolecular networks

Area of Science:

• Biomolecular networks
• Causal inference
• Systems biology

Background:

• Causal query estimation in biomolecular networks relies on valid adjustment sets to eliminate estimator bias.
• Multiple valid adjustment sets can exist, differing in variance, impacting estimation accuracy.
• Current graph-based methods for selecting adjustment sets in partially observed networks minimize asymptotic variance but overlook data generation processes.

Purpose of the Study:

• To develop a novel approach for deriving 'optimal adjustment sets' in biomolecular networks.
• To address the limitations of topology-based criteria that fail to distinguish variances when data generation processes differ.
• To improve the accuracy of causal effect estimation and intervention characterization.

Main Methods:

• Empirically learning data generating processes from historical experimental data.
• Characterizing estimator properties, including bias and finite-sample variance, through simulation.
• Developing an approach that considers data characteristics, estimator properties, and cost.

Main Results:

• Demonstrated the utility of the proposed approach in four diverse biomolecular case studies.
• Showcased improved selection of adjustment sets by accounting for data generation specifics.
• Provided an implementation and reproducible case studies for wider adoption.

Conclusions:

• The proposed method for deriving optimal adjustment sets enhances causal inference in biomolecular networks.
• Accounting for data generation processes leads to more robust and accurate estimation of intervention effects.
• This approach offers a significant improvement over existing graph-based methods, particularly in complex, partially observed biological systems.