Analysis of a Single Cell RNA-seq Workflow by Random Matrix Theory Methods.

Area of Science:

• Computational Biology
• Genomics
• Statistical Modeling

Background:

• Single-cell RNA sequencing (scRNAseq) generates complex count matrices.
• Existing clustering methods lack theoretical frameworks for explaining cluster formation or absence.
• Random matrix theory (RMT) has recently been applied to approximate scRNAseq count matrix eigenvalues.

Purpose of the Study:

• To extend RMT applications to the entire scRNAseq workflow.
• To develop theoretical tools for predicting scRNAseq clustering based on differential gene expression.
• To quantify differential expression and its impact on workflow predictions.

Main Methods:

• Modeling scaled scRNAseq count matrices using random matrices with normally distributed entries.
• Applying RMT to derive predictive formulas for scRNAseq workflow components, including clustering.
• Utilizing simulated and real scRNAseq datasets to validate RMT-based predictions.

Main Results:

• RMT-based predictions show accuracy under specific differential expression conditions.
• Real scRNAseq datasets often violate these conditions, introducing bias in predictions.
• The developed RMT approach provides more accurate predictions than naive estimators.

Conclusions:

• This work presents the first predictive formulas for scRNAseq workflows based on RMT.
• The findings highlight the potential of RMT for understanding and predicting scRNAseq data analysis outcomes.
• Future work is needed to refine predictions and address biases observed in real datasets.