Abstract
Quality control (QC) is a critical component of single-cell RNA-seq (scRNA-seq) processing pipelines. Current approaches to QC implicitly assume that datasets are comprised of one cell type, potentially resulting in biased exclusion of rare cell types. We introduce SampleQC, which robustly fits a Gaussian mixture model across multiple samples, improves sensitivity, and reduces bias compared to current approaches. We show via simulations that SampleQC is less susceptible to exclusion of rarer cell types. We also demonstrate SampleQC on a complex real dataset (867k cells over 172 samples). SampleQC is general, is implemented in R, and could be applied to other data types.