NG-SEM: an effective non-Gaussian structural equation modeling framework for gene regulatory network inference from...

Area of Science:

• Systems Biology
• Bioinformatics
• Computational Biology

Background:

• Gene regulatory network (GRN) inference is crucial in systems biology.
• Single-cell RNA sequencing (scRNA-seq) data presents unique challenges like dropouts and complex noise for GRN inference.

Purpose of the Study:

• To develop a more accurate method for GRN inference from scRNA-seq data.
• To address the limitations of traditional models in handling complex noise structures.

Main Methods:

• Extended the traditional structural equation modeling (SEM) framework.
• Incorporated a flexible noise modeling strategy using Gaussian mixtures.
• Utilized the Expectation-Maximization algorithm and weighted least-squares for optimization.
• Employed Akaike Information Criteria for selecting Gaussian mixture components.

Main Results:

• The proposed non-Gaussian SEM (NG-SEM) framework demonstrated improved performance over traditional Gaussian SEM on synthetic data.
• NG-SEM outperformed five state-of-the-art methods on real biological datasets.
• The Gaussian mixture approach effectively approximates complex noise structures in biological systems.

Conclusions:

• NG-SEM offers a robust and accurate approach for GRN inference from scRNA-seq data.
• The flexible noise modeling strategy is key to improving GRN inference accuracy.
• This method advances the analysis of gene regulatory mechanisms in complex biological systems.