IC2: interventional dynamical causality under latent confounders

Area of Science:

• Causal Inference
• Systems Biology
• Machine Learning

Background:

• Causal inference is crucial for understanding complex systems.
• Existing methods struggle with non-interventional data and latent confounders.
• There's a need for methods that infer interventional effects from observational data.

Purpose of the Study:

• To introduce a novel method, IC2, for deciphering interventional dynamical causality.
• To enable causal inference from non-interventional data, even with latent confounders.
• To overcome limitations of current association-based or intervention-requiring methods.

Main Methods:

• Theoretically grounded in the dual orthogonal decomposition theorem.
• Computationally implemented using deep neural networks to construct interventional data.
• Utilizes observed non-interventional data to infer causality.

Main Results:

• IC2 outperforms alternative methods in recovering causal structures.
• Validated by true interventional effects in knockout experiments.
• Successfully reconstructed biological networks and predicted CRISPR perturbation effects.

Conclusions:

• IC2 effectively estimates interventional effects from non-interventional data.
• The method is powerful for biological applications where direct intervention is difficult.
• IC2 advances causal inference in the presence of latent confounders.