dotears: Scalable and consistent directed acyclic graph estimation using observational and interventional data.

Area of Science:

• Computational Biology
• Systems Biology
• Genomics

Background:

• Inferring gene regulatory networks is crucial for understanding cellular mechanisms.
• Existing methods for learning causal structure from observational data face challenges like identifiability and sensitivity to error variance.
• New assays like Perturb-seq enable parallel CRISPR interventions linked to transcriptomic readouts, offering richer data for network inference.

Purpose of the Study:

• To develop a robust computational framework for inferring causal gene regulatory networks.
• To leverage both observational and interventional data for more accurate network reconstruction.
• To address limitations of existing score-based methods, particularly their sensitivity to error variance structure.

Main Methods:

• Introduced dotears, a continuous optimization framework for causal structure inference.
• Assumed a linear Structural Equation Model and exploited structural consequences of hard interventions.
• Developed a method to estimate and correct for error variance structure using interventional data.

Main Results:

• dotears is a provably consistent estimator of the true directed acyclic graph (DAG) under mild assumptions.
• Outperformed state-of-the-art methods in various simulation scenarios.
• In real biological data, dotears-inferred edges showed higher precision and recall, validated by differential expression tests and protein-protein interaction data.

Conclusions:

• dotears provides a powerful and accurate approach for inferring causal gene regulatory networks.
• The method effectively integrates observational and interventional data, improving upon existing techniques.
• dotears demonstrates significant potential for advancing systems biology research through precise network reconstruction.