Estimation of high-dimensional directed acyclic graphs with surrogate intervention.

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Directed acyclic graphs (DAGs) model causal relationships, typically requiring interventional data.
• High-dimensional complex systems often lack available interventional data.
• Estimating causal structure from observational data is desirable but challenging.

Purpose of the Study:

• To develop a Bayesian framework for estimating DAGs using surrogate interventional data.
• To address the limitations of inferring causal structure when direct interventions are not possible.
• To apply the method to expression quantitative trait locus (eQTL) studies.

Main Methods:

• Constructing a DAG skeleton using penalized regressions and partial correlation tests.
• Estimating posterior probabilities of edge directions by incorporating surrogate intervention data (DNA variations).
• The proposed method is named surrogate intervention recovery of a DAG (sirDAG).

Main Results:

• Demonstrated the utility of sirDAG through simulations.
• Successfully applied sirDAG to an eQTL study in 550 breast cancer patients.
• sirDAG effectively estimates causal structure from observational data with surrogate interventions.

Conclusions:

• The sirDAG method provides a robust approach for causal structure inference in complex biological systems.
• Surrogate interventions, like DNA variations in eQTL studies, can be leveraged to estimate causal DAGs.
• This framework advances the analysis of observational data for uncovering causal relationships.