Fig. 2

Single-cell RNA sequencing sensitivity. a Number of detected genes. Each point represents a single sample. Horizontal black lines indicate group mean. Boxes indicate ± 2 Sd.. Gray horizontal lines indicate 95% CI for the expected number of genes in a diluted replicate, assuming total (dark gray) or poly-A (light gray) RNA. See Methods . b Probability of gene detection as a function of the expected number of input molecules estimated using logistic model (see main text and Methods ). Horizontal lines indicate 50% and 90% probability. Vertical lines indicate 1 and at 4.605 molecules (99% probability of ≥ 1 molecule present in diluted replicate). Bands indicate 95% CI. Black line indicates probability of ≥ 1 molecule present in a diluted replicate. c Odds ratio for gene detection as a function of sequencing depth (+500,000 reads). Horizontal line indicates an odds ratio of one (no gain in detection sensitivity). Band indicates 95% CI. d Odds ratios for differences in biophysical trait values. Error bars indicate a 95% CI. “*” indicates significant difference across pairs of methods (Bonferonni corrected p < 0.05). The odds ratio for length and secondary structure are shown for the increase from 25%ile length (structure) in the HBR and UHR transcriptomes to 75%ile length (structure). e Odds ratio for an increase of 0.01 in GC content. Bands indicate 95% CI. f Boxplot of gene mappability, or the fraction of the gene body that can be aligned to uniquely (see Methods ) for computationally ambiguous gene detection outliers (wide boxes) and background genes (narrow boxes). Both undetected (Undet.) and detected (Det.) outliers are shown. “*” indicates significant difference (Wilcoxon rank-sum two-way test p < 0.05). g As in F, but for the fraction of the gene body that overlaps in genomic position with a separate gene annotation. h Nucleotide coverage. Observed over expected coverage normalized for expression level as a function of absolute 3' to 5' position. See Methods . i Comparison of nucleotide coverage with uniform distribution. Empirical CDF is of normalized per nucleotide coverage. Black diagonal line indicates uniformity. Kolmogorov-Smirnov (KS) statistic for difference from the uniform distribution is in the bottom right, with larger values indicating greater difference between the distributions. j Coverage for genes with different expression levels. Relative 5' to 3' coverage, calculated over 100 equally spaced bins for four expression level categories (rows). See Methods . Abbreviations: confidence interval (CI); Cumulative distribution function (CDF)