SCING: Inference of robust, interpretable gene regulatory networks from single cell and spatial transcriptomics

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Gene regulatory network (GRN) inference is crucial for understanding biological systems and diseases.
• Current single-cell RNA sequencing (scRNA-seq) methods for GRN inference have limitations in accuracy and speed.

Purpose of the Study:

• To introduce Single Cell INtegrative Gene regulatory network inference (SCING), a novel computational approach for robust GRN identification.
• To enhance the accuracy and interpretability of GRN inference using diverse transcriptomics data.

Main Methods:

• Developed SCING, a method combining gradient boosting and mutual information.
• Applied SCING to single-cell RNA sequencing (scRNA-seq), single-nuclei RNA sequencing (snRNA-seq), and spatial transcriptomics data.
• Validated SCING using Perturb-seq datasets, held-out data, and the mouse cell atlas.

Main Results:

• SCING demonstrated superior accuracy and biological interpretability compared to existing GRN inference methods.
• SCING successfully identified disease-specific subnetworks in human Alzheimer's disease and mouse models.
• The method inherently corrects for batch effects and reveals spatial transcriptomics insights.

Conclusions:

• SCING provides a robust and accurate platform for gene regulatory network inference across multiple transcriptomics modalities.
• SCING enhances understanding of cellular physiology and disease pathogenesis, particularly in complex conditions like Alzheimer's disease.