Compression of quantification uncertainty for scRNA-seq counts.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) quantification is challenged by multi-mapping reads, leading to underestimated gene counts.
• Existing methods often ignore these multi-mapping reads, introducing uncertainty.
• While inferential replicates capture this uncertainty, they increase computational demands.

Purpose of the Study:

• To develop an efficient method for handling quantification uncertainty in scRNA-seq data.
• To reduce storage and memory requirements associated with inferential replicates.
• To improve the accuracy of downstream statistical analyses, such as differential expression.

Main Methods:

• Implemented a 'compression' strategy, storing only mean and variance of inferential replicates.
• Generated 'pseudo-inferential' replicates from a negative binomial distribution.
• Extended the Swish method to incorporate pseudo-inferential replicates.

Main Results:

• Reduced disk storage to 9% and memory usage to 6% of original levels.
• Decreased false positive rates by over a third in trajectory-based differential expression analyses.
• Demonstrated no loss in performance when extending the Swish method with pseudo-inferential replicates.
• Highlighted significant underestimation of counts for important genes when discarding multi-mapping reads.

Conclusions:

• Compressing inferential replicates is sufficient to capture gene-level quantification uncertainty in scRNA-seq.
• The proposed method significantly reduces computational resources without compromising analytical performance.
• Accurate quantification, including handling multi-mapping reads, is crucial for reliable gene expression analysis.