A hybrid single cell demultiplexing strategy that increases both cell recovery rate and calling accuracy.
Article
Overview
abstract
UNLABELLED: Recent advances in single cell RNA sequencing allow users to pool multiple samples into one run and demultiplex in downstream analysis, greatly increasing the experimental efficiency and cost-effectiveness. However, the expensive reagents for cell labeling, limited pooling capacity, non-ideal cell recovery rate and calling accuracy remain great challenges for this approach. To date, there are two major demultiplexing methods, antibody-based cell hashing and Single Nucleotide Polymorphism (SNP)-based genomic signature profiling, and each method has advantages and limitations. Here, we propose a hybrid demultiplexing strategy that increases calling accuracy and cell recovery at the same time. We first develop a computational algorithm that significantly increases calling accuracy of cell hashing. Next, we cluster all single cells based on their SNP profiles. Finally, we integrate results from both methods to make corrections and retrieve cells that are only identifiable in one method but not the other. By testing on several real-world datasets, we demonstrate that this hybrid strategy combines advantages of both methods, resulting in increased cell recovery and calling accuracy at lower cost. HIGHLIGHTS: An improved algorithm for cell hashing that distinguishes true positive from background for each individual hashtag at higher accuracyThis hybrid strategy increases cell recovery and calling accuracy while lowering experimental costThis hybrid demultiplexing strategy is applicable for single-cell RNA sequencing with different donor species, subjects, and cell populationsDoublet rate is a major determinant of the performance of SNP-based demultiplexing method.
