scIBD: a self-supervised iterative-optimizing model for boosting the detection of heterotypic doublets in single-cell...

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Droplet-based microfluidics are crucial for single-cell sequencing.
• The presence of doublets, where two cells are captured as one, introduces significant bias in downstream analyses.
• Existing doublet detection methods face challenges, particularly for single-cell chromatin accessibility sequencing (scCAS) data.

Purpose of the Study:

• To develop an effective method for detecting heterotypic doublets in scCAS data.
• To address the limitations of current doublet detection approaches.
• To improve the accuracy and reliability of single-cell chromatin accessibility analyses.

Main Methods:

• Proposed scIBD, a self-supervised iterative-optimizing model.
• Introduced an adaptive strategy for simulating high-confidence heterotypic doublets.
• Employed iterative optimization for self-supervised doublet detection.

Main Results:

• scIBD demonstrated superior performance and robustness across various simulated and real datasets.
• Comprehensive benchmarking confirmed the outperformance of scIBD compared to existing methods.
• Downstream biological analyses validated the efficacy of doublet removal using scIBD.

Conclusions:

• scIBD effectively detects and removes heterotypic doublets in scCAS data.
• The proposed method enhances the accuracy of single-cell chromatin accessibility analyses.
• scIBD offers a robust solution for mitigating doublet-induced bias in scCAS studies.