A Bayesian framework for inter-cellular information sharing improves dscRNA-seq quantification.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Accurate gene-level abundance estimation is critical for droplet-based single-cell RNA sequencing (dscRNA-seq) data analysis.
• Challenges in dscRNA-seq data include sparsity, 3' bias, and multi-mapping reads, complicating gene quantification.
• Disambiguating reads for genes with no unique mapping is a significant hurdle in pre-processing dscRNA-seq data.

Purpose of the Study:

• To develop a Bayesian framework for improved gene quantification in dscRNA-seq data.
• To enhance the estimation of gene-level abundances, especially for genes with multi-mapping reads.
• To provide a principled method for information sharing across cells and data modalities.

Main Methods:

• Implemented a Bayesian framework for information sharing across cells and multiple data modalities.
• Utilized an anchor-based approach to connect cells with similar gene expression patterns.
• Learned empirical priors to improve alevin's gene multi-mapping resolution algorithm.

Main Results:

• The new model significantly improves per-cell gene-level estimates for dscRNA-seq data.
• Successfully addressed quantification challenges for genes lacking uniquely mapping reads.
• Demonstrated the framework's effectiveness on both simulated and real datasets.

Conclusions:

• The developed Bayesian framework enhances gene quantification accuracy in dscRNA-seq.
• Information sharing across cells and modalities offers a robust approach to handle data limitations.
• The method provides a valuable tool for more precise single-cell transcriptomic analysis.