MetaSEM: Gene Regulatory Network Inference from Single-Cell RNA Data by Meta-Learning

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene regulatory networks (GRNs) are essential for understanding cell states.
• Inferring GRNs from single-cell RNA sequencing (scRNA-seq) data faces challenges like high dimensionality, sparsity, and limited labeled data.

Purpose of the Study:

• To develop a robust meta-learning framework for GRN inference to identify critical regulatory factors.
• To address data challenges in scRNA-seq data, including high dimensionality, sparsity, and scarcity of labels.

Main Methods:

• A meta-learning framework was employed to optimize parameters for high-dimensional, sparse data.
• A few-shot learning approach was utilized to overcome the lack of labeled data.
• A structural equation model (SEM) was integrated to identify key regulators, with parameter optimization embedded in bi-level optimization for robustness on small datasets.

Main Results:

• The proposed model effectively extracts features consistent with GRN reasoning, demonstrating robustness on small-scale datasets.
• Identified regulators showed a strong correlation with gene expression specificity.
• The inferred GRNs highlighted important regulators for cell type identification.

Conclusions:

• The meta-learning framework successfully captures regulatory factors in single-cell GRN inference.
• The identified regulators are vital for accurate cell type recognition, as confirmed by visualization results.