demuxmix: demultiplexing oligonucleotide-barcoded single-cell RNA sequencing data with regression mixture models

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Droplet-based single-cell RNA sequencing (scRNA-seq) enables large-scale transcriptome analysis.
• Pooling samples with hashtag oligonucleotides (HTOs) reduces costs and batch effects.
• Accurate demultiplexing of HTOs is critical but challenging due to noise and low-quality data.

Purpose of the Study:

• Introduce demuxmix, a novel demultiplexing method for scRNA-seq.
• Improve accuracy and reliability of sample assignment in pooled scRNA-seq experiments.
• Provide probabilistic error estimates for droplet assignments.

Main Methods:

• Developed demuxmix, a method utilizing negative binomial regression mixture models.
• Implemented a probabilistic classification framework to quantify assignment uncertainty.
• Leveraged the association between RNA expression and HTO counts to refine demultiplexing.

Main Results:

• demuxmix demonstrates improved performance over existing methods on real and simulated data.
• Probabilistic framework allows for identification and removal of uncertain droplet assignments.
• Successfully demonstrated demultiplexing of pooled labeled and unlabeled cells.

Conclusions:

• demuxmix offers a robust and accurate approach for HTO-based demultiplexing in scRNA-seq.
• The method enhances data quality by providing error probabilities and improving assignment accuracy.
• Facilitates more reliable analysis of pooled samples in complex experimental designs.