Quantification errors and biases. (A) Scatter plot of read counts for each ERCC transcript in two different libraries of human RNA-seq with 2% ERCC spike-ins (K562 A+ Repl.1 and K562 A+ Repl.2, libraries 7, 8). (B) Scatter plot of fold deviation between replicates versus read counts for a given ERCC RNA. (C,D) Read counts for two example ERCCs relative to the total number of ERCC reads across 44 different libraries (libraries 7–50) with ERCC spike-in H. sapiens RNA samples (black line), the negative binomial distribution (solid gray), and random samples (n = 44) from the negative binomial distribution (dotted gray). The observed distribution fits a negative binomial model over a Poisson model (P < 2.2 × 10−16, likelihood ratio). (E–G) Scatter plots of the fold deviation between observed and expected read count for each ERCC in the 100% ERCC library (library 6) compared with read count (E), GC content (F), and ERCC RNA length (G).
