ESCO: single cell expression simulation incorporating gene co-expression.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Gene-gene co-expression networks (GCN) are crucial for understanding gene interactions.
• Single-cell RNA sequencing enables detailed analysis of co-expression within cell types.
• Existing simulators often lack robust gene co-expression simulation capabilities.

Purpose of the Study:

• To introduce ESCO, a novel simulator designed to accurately model gene co-expression.
• To evaluate the performance of imputation methods for GCN recovery using simulated single-cell data.
• To provide a tool for benchmarking imputation and denoising methods in single-cell genomics.

Main Methods:

• Developed ESCO, a simulator incorporating copula-based gene co-expression modeling.
• Preserved marginal gene expression simulation accuracy from existing methods.
• Assessed imputation method performance on GCN recovery using ESCO-generated data.

Main Results:

• Imputation generally improves GCN recovery in non-sparse single-cell data.
• Ensemble imputation methods demonstrated superior performance in GCN recovery.
• Data aggregation methods are preferable when dealing with excessive zero counts.

Conclusions:

• ESCO effectively simulates gene co-expression, aiding in the assessment of computational methods.
• The findings provide guidance on selecting appropriate methods for GCN recovery based on data sparsity.
• Results were validated using both mouse and human brain cell datasets.