Inferring Cell-Type-Specific Co-Expressed Genes from Single Cell Data.

Area of Science:

• Genomics
• Computational Biology
• Single-cell analysis

Background:

• Cell-type-specific gene co-expression networks are vital for understanding gene relationships.
• Current methods for inferring these networks from single-cell data often lack robust false positive control.
• Inadequate false positive control can lead to misleading conclusions about method performance, even with high reproducibility or functional coherence.

Purpose of the Study:

• To develop an efficient simulation tool for deriving empirical p-values to control false positives in co-expression inference.
• To assess the performance of p-value-based approaches for inferring cell-type-specific co-expressions using simulated and real single-cell data.
• To address biases in method assessment, including random overlaps and expression level biases in known biological networks.

Main Methods:

• Development of a simulation tool to generate empirical p-values for gene co-expression inference.
• Evaluation of p-value-based methods using both simulated and real single-cell datasets.
• Analysis of potential biases in network comparisons, such as random overlap and expression level effects.

Main Results:

• The simulation tool effectively derives empirical p-values to control false positives in co-expression inference.
• The p-value-based approach demonstrates power in inferring cell-type-specific co-expressions.
• Identified and illustrated the impact of biases, including random overlaps and expression level bias, on method assessment.

Conclusions:

• Controlling false positives is essential for reliable gene co-expression network inference.
• The proposed simulation-based p-value method provides a robust strategy for evaluating co-expression inference tools.
• This approach enhances the accuracy and trustworthiness of findings derived from single-cell data analysis.