demuxSNP: supervised demultiplexing single-cell RNA sequencing using cell hashing and SNPs

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Multiplexing single-cell RNA sequencing (scRNA-seq) reduces costs but faces challenges with cell hashing quality and class imbalance.
• Existing demultiplexing algorithms often rely on a single data modality, limiting their performance.

Purpose of the Study:

• To introduce demuxSNP, a supervised algorithm that integrates cell hashing and single-nucleotide polymorphism (SNP) data for robust scRNA-seq demultiplexing.
• To address the limitations of existing methods, particularly in scenarios with low hashing quality and imbalanced datasets.

Main Methods:

• demuxSNP utilizes probabilistic hashing for initial cell assignment and infers genotypes of singlet/doublet clusters.
• A nearest-neighbor approach adapted for missing data is employed to classify uncertain or negative cells.
• The algorithm was benchmarked against standalone hashing, genotype-free SNP, and hybrid methods using simulated and real renal cell carcinoma data.

Main Results:

• demuxSNP demonstrated superior performance over standalone hashing methods, especially with low-quality hashing data.
• The algorithm improved overall classification accuracy and recovered more high-quality RNA cells.
• Unlike genotype-free SNP and hybrid methods, demuxSNP's supervised approach showed greater robustness to varying doublet rates and class size imbalance.

Conclusions:

• demuxSNP effectively demultiplexes scRNA-seq datasets by combining hashing and SNP data, proving advantageous for samples with genetic distinctions and low hashing quality.
• The method enhances data utility by recovering unassigned or negative cells with high RNA quality.
• Associated data simulation and benchmarking pipelines, along with the demuxSNP R/Bioconductor package, are publicly available.