Recovering Gene Interactions from Single-Cell Data Using Data Diffusion.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity.
• Technical noise, particularly "dropout" (under-sampling of mRNA), limits scRNA-seq data quality and obscures gene-gene relationships.
• Accurate analysis of scRNA-seq data is essential for discovering cellular states and regulatory networks.

Purpose of the Study:

• To develop a computational method for denoising scRNA-seq data and imputing missing gene expression values.
• To address the challenge of technical noise and improve the recovery of true biological signals.
• To enable the discovery of gene-gene relationships and cellular continua from noisy scRNA-seq data.

Main Methods:

• Developed MAGIC (Markov affinity-based graph imputation of cells), a novel imputation method.
• Employed data diffusion to share information across similar cells, effectively denoising the cell count matrix.
• Applied the method to various biological systems, including the epithelial-to-mesenchymal transition.

Main Results:

• MAGIC effectively denoises scRNA-seq data and imputes missing transcripts, recovering gene-gene relationships.
• The method reveals a phenotypic continuum in epithelial-to-mesenchymal transition, highlighting intermediate stem-like cell states.
• MAGIC successfully infers known and novel regulatory interactions without requiring experimental perturbations.

Conclusions:

• MAGIC is a robust computational tool for enhancing scRNA-seq data quality and uncovering biological insights.
• The method facilitates the identification of complex cellular states and regulatory networks from noisy single-cell data.
• MAGIC demonstrates the potential for computational approaches to reveal biological mechanisms from observational scRNA-seq data.