Genetics

PACER: Acyclic Causal Discovery from Large-Scale Interventional Data

Ramon Viñas Torné, Sílvia Fàbregas Salazar, Soyon Park, Ivo Alexander Ban, Artyom Gadetsky, Nikita Doikov, Maria Brbić

Featured May 26, 2026

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

This method builds causal graphs directly to avoid cycles, using a clever way to order variables and pick connections, making it super fast and accurate for finding relationships in huge datasets.

In depth

The paper introduces PACER, a scalable framework for causal discovery that guarantees acyclicity by design. It achieves this by defining a distribution over Directed Acyclic Graphs (DAGs) through a joint model of variable permutations (using a Plackett-Luce distribution) and edge probabilities. For linear-Gaussian mechanisms, the authors derive a closed-form expression for the expected interventional log-likelihood, yielding substantial computational gains and enabling scalability to thousands of variables.

Key Takeaways

1
PACER guarantees acyclicity by design by directly modeling a distribution over DAGs using variable permutations and edge probabilities, avoiding soft acyclicity penalties.
2
The framework provides a unified likelihood-based objective that jointly leverages observational and interventional data, supporting flexible conditional density models and structural prior knowledge.
3
For linear-Gaussian mechanisms, PACER derives an exact closed-form expression for the expected interventional log-likelihood, leading to significant computational speedups (up to two orders of magnitude) and $O (n^{2})$ scalability.

Conceptual Flow

HIGH LEVEL

Methodology: Building Acyclic Causal Graphs

Instead of guessing graphs and then fixing cycles, this method builds graphs that are guaranteed to have no cycles from the start.

Variable Orderings

Edge Choices

Combine to Form

Valid Causal Graph

Results: Fast and Accurate Discovery

By building graphs smartly, the method finds correct relationships much faster than old ways, even for very large systems.

Old Methods (Slow, Unstable)

New Method (Fast, Stable)

Process Data

Accurate Causal Map