deMULTIplex2: robust sample demultiplexing for scRNA-seq.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell sample multiplexing enhances scRNA-seq throughput and reduces costs by tagging cells with sample-specific barcodes.
• Accurate computational assignment of cell-to-sample barcodes is crucial but challenging with large, imbalanced, or noisy datasets.
• Existing methods struggle with real-world experimental data complexities, including cross-contamination of sample tags.

Approach:

• Introduced deMULTIplex2, a mechanism-guided classification algorithm for multiplexed scRNA-seq data.
• Developed a statistical model based on the physical mechanism of tag cross-contamination.
• Employed generalized linear models and expectation-maximization for probabilistic cell sample inference and singlet classification.

Key Points:

• deMULTIplex2 significantly improves cell recovery across diverse and challenging multiplexed scRNA-seq datasets.
• The algorithm demonstrates superior performance, particularly with large-scale, noisy data, and imbalanced sample compositions.
• Model fit validated using simulated and real-world datasets via Randomized Quantile Residuals.

Conclusions:

• deMULTIplex2 offers a robust solution for accurate cell demultiplexing in scRNA-seq.
• The method outperforms current algorithms in complex experimental scenarios.
• Enables more reliable and comprehensive analysis of multiplexed single-cell data.