Identifying gene expression programs in single-cell RNA-seq data using linear correlation explanation.

Area of Science:

• Genomics and bioinformatics
• Computational biology
• Molecular biology

Background:

• Single-cell RNA sequencing (scRNA-seq) advances gene regulation studies.
• Current methods primarily identify cell type-specific gene expression programs (GEPs).
• Characterizing GEPs for biological processes and stimuli responses is limited.

Purpose of the Study:

• To infer biologically meaningful GEPs from scRNA-seq data.
• To associate GEPs with cellular phenotypes and activity programs.
• To develop a robust method for analyzing complex scRNA-seq signals.

Main Methods:

• Applied linear CorEx, a machine learning approach, to group genes based on total correlation optimization.
• Utilized simulated and real-world scRNA-seq datasets for GEP inference.
• Employed transfer learning to project inferred GEPs across datasets.

Main Results:

• Linear CorEx outperformed similar methods in identifying cell types and activity programs on simulated data.
• Identified biologically relevant GEPs in mouse dentate gyrus and embryonic colon development data.
• Demonstrated cross-species sensitivity and transfer learning potential of linear CorEx.

Conclusions:

• Linear CorEx is a valuable tool for comprehensive scRNA-seq data analysis.
• Provides deeper insights into gene expression dynamics and cellular heterogeneity.
• Enhances understanding of regulatory mechanisms in complex biological systems.