A unified model-based framework for doublet or multiplet detection in single-cell multiomics data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Droplet-based single-cell sequencing assumes one cell per droplet for accurate omics profiling.
• Multiplets (multiple cells in one droplet) lead to incorrect cell type annotations and obscure biological insights.
• Multiplets are particularly problematic in single-cell multiomics, where integrating data modalities can aggregate multiplet clusters.

Purpose of the Study:

• To develop a robust computational framework for detecting and mitigating multiplets in single-cell multiomics data.
• To address the challenge of erroneous cell type annotations caused by multiplets in integrated multiomics datasets.
• To provide a method that effectively handles cross-modality multiplet signals.

Main Methods:

• Proposed a compound Poisson model-based framework for multiplet detection.
• Utilized experimental cell hashing data as ground truth for validating multiplet status.
• Conducted trimodal DOGMA-seq experiments generating 17 benchmarking datasets (280,123 droplets).

Main Results:

• The compound Poisson model effectively detects multiplets in single-cell multiomics data.
• The proposed method successfully integrates cross-modality multiplet signals.
• Demonstrated superior performance compared to single-omics methods in eliminating multiplet clusters.

Conclusions:

• The developed compound Poisson model is essential for accurate single-cell multiomics analysis.
• This framework enhances the reliability of cell type annotation by effectively removing multiplet-derived artifacts.
• The method overcomes limitations of existing approaches in handling multiplets within multiomics data.