Supplementary Figure 3: Direct capture Perturb-seq performs comparably to GBC Perturb-seq for phenotypic analysis.
From: Combinatorial single-cell CRISPR screens by direct guide RNA capture and targeted sequencing
a) Mean target knockdown (fraction mRNA remaining) for each targeting guide (n=30) in the indicated experiments. For each guide, the data point represents the mean normalized expression level of the target gene across cells bearing the corresponding guide divided by the mean normalized expression level of the target gene in control cells (NegCtrl3). b) The Pearson correlation of pseudo-bulk expression profiles from direct capture Perturb-seq and GBC Perturb-seq experiments for each perturbation (n=30 targeting guides). Profiles were generated from the top 100 most differentially expressed genes in GBC Perturb-seq. Grey lines indicate medians. c) Scatterplot indicates the relationship between the number of differentially expressed genes for each guide (determined by a two-sided, two-sample Kolmogorov-Smirnov test using GBC Perturb-seq data) and the Pearson correlation of pseudo-bulk expression profiles between GBC Perturb-seq and direct capture Perturb-seq on the indicated platform (n=30 targeting guides per platform). d) Scatterplots of the balanced accuracy of random forest classifiers trained to distinguish perturbed and unperturbed (NegCtrl3) cells for each of n=30 targeting guides on the indicated platforms. Direct capture Perturb-seq accuracies were highly correlated with GBC Perturb-seq (Pearson correlation: r=0.91 for 3’ sgRNA-CR1cs1 capture; r=0.90 for 5’ sgRNA-CR1 capture). We failed to detect significant differences in performance between direct capture Perturb-seq and GBC Perturb-seq (Wilcoxon signed-rank two-sided test: p=0.2 for 3’ sgRNA-CR1cs1; p=0.6 for 5’ sgRNA-CR1 capture).
