Skip to main content
Fig. 3 | Genome Biology

Fig. 3

From: pipeComp, a general framework for the evaluation of computational pipelines, reveals performant single cell RNA-seq preprocessing tools

Fig. 3

Effect of filtering on cell subpopulation structure and clustering. a Filtering on the basis of distance to the whole distribution can lead to strong bias against certain subpopulations. The dashed line indicates a threshold of 2.5 median absolute deviations (MADs) from the median of the overall population. b Relationship between the maximum subpopulation exclusion rate and the average clustering accuracy per subpopulation across various filtering strategies.The datasets for which the presence of doublets could be confirmed with SNP genotypes are labeled in bold. Of note, doublet removal appears to have a neutral or positive impact even when, due to the design, there are no heterotypic doublets in the data. The PCA methods refer to multivariate outlier detected as implemented in scater (see the “Methods” section for details)

Back to article page