Estimating causal effects with a non-paranormal method for the design of efficient intervention experiments

Area of Science:

• Genomics and Systems Biology
• Computational Biology and Bioinformatics
• Statistical Genetics

Background:

• Gene knockdown and overexpression are crucial for identifying genes impacting cellular functions.
• High-throughput sequencing necessitates robust methods for identifying genes driving phenotypic changes.
• Conventional causal inference methods often fail due to the assumption of normality.

Purpose of the Study:

• To introduce a non-paranormal method for testing conditional independence in causal inference.
• To develop the non-paranormal intervention-calculus (NPN-IDA) for estimating gene-phenotype relationships without a directed acyclic graph (DAG).
• To enable accurate causal gene ranking by incorporating cumulative effects through cascaded pathways.

Main Methods:

• Developed a non-paranormal approach to relax Gaussian assumptions in the PC-algorithm for DAG estimation.
• Introduced NPN-IDA, a causal inference framework for scenarios lacking a predefined DAG.
• Applied causal inference with the non-paranormal method for estimating DAGs and identifying causal genes.

Main Results:

• The non-paranormal method significantly enhanced DAG estimation accuracy on synthetic data compared to the standard PC-algorithm.
• NPN-IDA demonstrated superior performance in identifying flowering time regulators in Arabidopsis thaliana.
• The method successfully identified regulators of white adipocyte browning in mice, highlighting its broad applicability.

Conclusions:

• The proposed non-paranormal method facilitates the design of efficient intervention experiments.
• This generalizable approach has wide-ranging applications, including drug discovery and understanding complex biological systems.
• Improved DAG estimation directly contributes to more accurate causal effect estimations in biological research.