Figure 4.

Patch-seq experimental confounds affect the numbers of genes detected per cell. (A) Schematic illustrating how spike-in mRNAs can be used to estimate how much mRNA was extracted per cell. (B) Violin plots showing numbers of protein-coding genes detected per cell across patch-seq datasets or the Ndnf subset of the Tasic dissociated-cell dataset. (C) Technical factors associated with numbers of genes detected per cell across datasets. Bars show standardized beta model coefficients with y-axis in units of standard deviations, allowing direct comparison of effects across factors and across datasets. Error bars indicate coefficient standard deviations. Positive (negative) model coefficients indicate technical factor is correlated with increased (decreased) detected gene counts per cell. Regression models calculated using only cells containing mRNA spike-ins. (D–H) Examples of univariate relationships between technical factors and detected gene count per cell (dots) across patch-seq datasets. Gray line shows best fit line. (D) Library size (count of sequenced reads per cell). (E) Spike-ins as a fraction of all sequenced reads per cell. Samples with lower cellular mRNA content (indicated by higher spike-in ratios) have lower gene counts. (F) Unmapped ratio, calculated as the ratio of exon-mapping reads to all sequenced reads (excluding spike-ins). (G) Cellular contamination index, quantified by summing normalized contamination scores across tested cell types (arbitrary units). (H) Overall percent variance explained by each dataset-specific statistical model shown in (C).