miQC: An adaptive probabilistic framework for quality control of single-cell RNA-sequencing data
Fig 1
Uniform and data-driven quality control (QC) thresholds for scRNA-seq data.
Cells (N = 6618) from one high-grade serous ovarian cancer (HGSOC) tissue sample (Sample ID: 16030X4) with the number of unique genes detected (x-axis) and percent of cell counts mapping to mitochondrial (mtDNA) genes (y-axis). (A) Illustration of removing cells with a uniform QC threshold of greater than 10% cell counts mapping to mtDNA genes (solid black line) and a more data-driven threshold of greater than 3 median absolute deviations (MADs) of the percent of counts mapping to mtDNA genes (dotted black line). (B) Using our data-driven approach (miQC), we fit a finite mixture of standard linear regression models with two lines (black lines) to calculate a posterior probability of being a compromised cell. (C) Cells shaded by their posterior probability of being compromised. (D) Discarding all cells with ≥ 75% probability of being compromised creates a data-driven QC threshold for scRNA-seq data.