SEMbap: Bow-free covariance search and data de-correlation.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene expression studies face challenges from biological and technical variations.
• Unobserved confounding variables hinder causal relationship discovery in high-dimensional data.
• Existing deconfounding methods have limitations in accuracy and efficiency.

Purpose of the Study:

• To develop a robust two-stage deconfounding procedure for gene expression data.
• To introduce the SEMbap() method based on Bow-free Acyclic Paths (BAP) search within Structural Equation Models (SEM).
• To evaluate SEMbap()'s performance against established deconfounding techniques.

Main Methods:

• SEMbap() employs a two-stage approach: BAP search using Shipley d-separation tests and fitting a Constrained Gaussian Graphical Model (CGGM) or using Graph Laplacian Principal Component Analysis (gLPCA).
• Exhaustive search for missing edges with significant covariance in the first stage.
• Obtaining a low-dimensional representation of bow-free edges structure in the second stage.

Main Results:

• SEMbap() accurately identifies hidden confounding variables in simulated and observed gene expression data.
• The BAP search approach demonstrates superior performance compared to four popular deconfounding methods.
• SEMbap() effectively controls the false positive rate while achieving good fitting and perturbation metrics.

Conclusions:

• SEMbap() offers a significant advancement in deconfounding methods for large-scale gene expression studies.
• The BAP search algorithm provides a reliable approach for uncovering hidden confounders.
• This method enhances the accuracy and reliability of causal inference in genomics research.