BAYESIAN DIFFERENTIAL CAUSAL DIRECTED ACYCLIC GRAPHS FOR OBSERVATIONAL ZERO-INFLATED COUNTS WITH AN APPLICATION TO...

Area of Science:

• Genomics and bioinformatics
• Statistical modeling
• Causal inference

Background:

• Observational count data often exhibit excessive zeros, common in genomics.
• Existing methods for learning causal networks (Directed Acyclic Graphs - DAGs) struggle with zero-inflated data.
• Identifying differential causal networks between experimental groups is vital for comparative studies.

Purpose of the Study:

• To propose a novel Bayesian differential zero-inflated negative binomial DAG (DAG0) model.
• To address the limitations of current methods in modeling zero-inflated count data and identifying network differences.
• To ensure causal relationships are identifiable from observational, cross-sectional data.

Main Methods:

• Development of the Bayesian differential zero-inflated negative binomial DAG (DAG0) model.
• Theoretical proof of identifiability for causal relationships from observational data.
• Application of parallel-tempered Markov chain Monte Carlo for Bayesian inference.

Main Results:

• The proposed DAG0 model effectively accounts for zero-inflation in count data.
• Causal relationships are proven to be fully identifiable from observational data.
• Simulations show superior performance compared to existing methods.
• Application to single-cell RNA-sequencing data yields biologically relevant insights.

Conclusions:

• The DAG0 model provides a robust framework for causal network inference with zero-inflated count data.
• The model facilitates the identification of differential causal structures between experimental groups.
• The identifiability proof offers a general technique applicable beyond the DAG0 model.