Modeling group heteroscedasticity in single-cell RNA-seq pseudo-bulk data.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Pseudo-bulk single-cell RNA-seq data often exhibits group heteroscedasticity.
• Unequal group variances can impede the accurate detection of differentially expressed genes.
• Standard bulk RNA-seq methods typically assume equal variances across groups.

Purpose of the Study:

• To introduce novel statistical approaches for analyzing pseudo-bulk single-cell RNA-seq data with heteroscedastic groups.
• To improve the detection of differentially expressed genes in the presence of unequal group variances.
• To provide robust methods that enhance both error control and statistical power.

Main Methods:

• Development of two new methods: voomByGroup and voomWithQualityWeights using a blocked design (voomQWB).
• These methods explicitly account for heteroscedasticity in group variances.
• Evaluation through simulations and experimental datasets.

Main Results:

• voomByGroup and voomQWB demonstrate superior performance compared to existing methods that ignore group heteroscedasticity.
• The proposed methods offer improved error control, reducing false positives.
• Enhanced statistical power leads to better detection of true differentially expressed genes.

Conclusions:

• The novel methods voomByGroup and voomQWB effectively address group heteroscedasticity in pseudo-bulk single-cell RNA-seq data.
• These approaches represent significant advancements for differential gene expression analysis in such datasets.
• Researchers can achieve more reliable and powerful results by employing these methods when group variances are unequal.