CMDdemux: an efficient single cell demultiplexing method.

Area of Science:

• Single-cell genomics
• Computational biology
• Bioinformatics

Background:

• Multiplexing technologies, like hashtag-based labeling, enable high-throughput single-cell sequencing by pooling cells from different donors.
• While beneficial, limitations in hashtag labeling can lead to low-quality data, challenging existing demultiplexing methods.
• Current demultiplexing tools struggle with low-quality datasets, hindering accurate cell identity assignment.

Purpose of the Study:

• To develop a robust demultiplexing method, CMDdemux, capable of handling low-quality multiplexing data.
• To improve the accuracy of distinguishing singlets, doublets, and negatives in pooled single-cell sequencing.
• To provide visualization tools for inspecting potentially misclassified droplets.

Main Methods:

• CMDdemux integrates hashtag and messenger RNA data for demultiplexing.
• Key steps include within-cell centered log-ratio (CLR) normalization, K-medoids clustering, and Mahalanobis distance-based classification.
• The method incorporates visualization tools for user inspection of droplet classifications.

Main Results:

• CMDdemux demonstrates high accuracy in distinguishing singlets, doublets, and negatives.
• Benchmarking shows CMDdemux consistently outperforms existing methods on both high- and low-quality datasets.
• The method exhibits robust performance on diverse low-quality multiplexing data, even where other methods fail.

Conclusions:

• CMDdemux offers a significant advancement in demultiplexing low-quality single-cell sequencing data.
• The method's robustness makes it effective across various multiplexing technologies and data qualities.
• CMDdemux enhances the reliability and throughput of pooled single-cell experiments.