DCI

DCI infers differences between causal gene regulatory networks from gene expression data to identify condition-specific changes in causal relationships.

Key Features:

• Efficient learning: DCI directly infers changes between two causal graphs derived from gene expression data across conditions, identifying edges that appear, disappear, or change in weight.
• Resource efficiency: The algorithm focuses on differences between graphs rather than estimating each large causal graph separately, improving sample and computational efficiency.
• Stability selection for robustness: The method uses stability selection across tuning parameters to determine a robust difference causal graph.
• Implementation: Implemented in the causaldag package with a Python implementation.

Scientific Applications:

• Gene regulatory network modeling: Inferring condition-specific causal graphs to support modeling and design of interventions that control gene regulation.
• Single-cell RNA-seq data analysis: Application to single-cell RNA sequencing data to explore gene regulatory mechanisms across conditions and cell states.
• Intervention prediction: Anticipating and validating outcomes of genetic or environmental interventions by comparing inferred difference graphs to known intervention effects.

Methodology:

DCI directly infers differences between two causal graphs from distinct gene expression datasets, identifies edge additions, deletions, and weight changes, focuses estimation on graph differences to improve sample and computational efficiency, and applies stability selection across tuning parameters; it is implemented in the causaldag package (Python).