Inferring gene regulatory networks from single-cell multiome data using atlas-scale external data.

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Gene regulatory network (GRN) inference traditionally uses gene expression data or low-resolution bulk data.
• Integrating chromatin accessibility and RNA sequencing data presents challenges due to limited independent data points for learning complex mechanisms.

Purpose of the Study:

• To develop a machine-learning method, LINGER (Lifelong Neural Network for Gene Regulation), for inferring GRNs from single-cell paired gene expression and chromatin accessibility data.
• To leverage atlas-scale external bulk data and transcription factor motif prior knowledge for enhanced GRN inference.

Main Methods:

• LINGER utilizes single-cell multiome data (gene expression and chromatin accessibility).
• Incorporates external bulk data and transcription factor motif information as manifold regularization.
• Applies a lifelong neural network approach for continuous learning and adaptation.

Main Results:

• LINGER demonstrates a fourfold to sevenfold relative increase in accuracy compared to existing methods.
• Reveals a complex regulatory landscape relevant to genome-wide association studies (GWAS).
• Enables enhanced interpretation of disease-associated variants and genes.

Conclusions:

• LINGER provides a powerful tool for GRN inference from single-cell multiome data.
• Facilitates the estimation of transcription factor activity from gene expression data for identifying driver regulators in case-control studies.
• Enhances understanding of disease mechanisms by linking genetic variants to regulatory elements and genes.