Extended Data Fig. 1: Quality control, scoring and clustering for TP53 Perturb-Seq experiment.

a. Distribution of lengths (in kb) of coding regions for 309 actionable genes from the Foundation Medicine Panel. The principal transcript as defined by the APPRIS database was used for each gene. b. Variant representation in the library. Number of barcode reads (y axis) for each tested variant (x axis), either after transduction and 2-day puromycin selection (‘no recovery’), or 42.5 hours after puromycin selection (‘42.5h recovery’). We refer to variants that are present in the pooled library as passing quality control in the main text. c-g. Quality control metrics. c. Cumulative distribution function (CDF) of number of cells (x axis) profiled for each variant, considering either all cells (gray) or only cells with a single variant (black). d. Distribution of the number of variants detected per cell. e. Distribution of the number of variant barcode (vbc) UMIs per cell per variant. f. The number of cells detected per variant (y axis) and the variant’s barcode expression (x axis, transcripts per 10,000 UMIs/cell (TP10K)) for cells with a single variant, colored by class (black: WT-like, light blue: Impactful I, dark blue: Impactful II). g. Distribution of mean variant barcode expression (TP10K, x axis). Variants with a fold change higher than 1.5 compared to the WT barcode are colored by variant class. h. sc-eVIP scores are independent of variant expression. Variant expression (y axis, TP10K) for variant (dots) with different sc-eVIP scores (x axis). i. Sensitivity for identifying impactful variants at an FDR of 5% (y axis), as a function of the number of principal components used (colors) and as a function of the number of cells per variant (x axis). Boxplots represent the results of 10 subsampling iterations, and show the median, with their ends representing the 25% and 75% quartiles, and with whiskers extending between (25% quartile - 1.5 interquartile range) and (75% quartile + 1.5 interquartile range) or the most extreme values in the data, if they fall within this range. j. Impact of number of cells on sc-eVIP scores. sc-eVIP scores (y axis) for each variant (x axis) computed with varying numbers of subsampled cells (rows), colored by variant class. Mean scores (center) and 1 standard deviation (error bars) are based on 10 subsampling iterations, and are colored by variant classs. k. Low-dimensional embedding of mean expression profiles of variants (dots), colored by variant class (left), or by Louvain clustering (right).