Fig. 7

Comparison of accuracy as measured by concordance to qRT-PCR on real RNA-seq data. Pearson correlation between RNA-seq-based gene expression level estimates (log(TPM*τ + 1)) and qRT-PCR-based measurements (ΔCt), across independently performed experiments. τ was used to maximize the correlation to adjust for any scale effect of the pseudocount, which allowed inclusion of zero estimates. Six RNA-seq data sets (UHRR-1 to −4, HBRR-1, MKN-28) and four qRT-PCR sets (UHRR MAQC [19], HBRR MAQC [19], UHRR Wang et al. [20], MKN-28 from Lee et al. [5]) were compared. RNA-seq-based quantification was run on two independent transcriptome models (ENSEMBL (E) and RefSeq (R)). The results from Lee et al. on an older version of RefSeq model (R’) was shown to the right for comparison (grey box). Gene-level estimates were obtained by summing the relevant isoform-level estimates, except for NEUMA, for which gene-level estimates were obtained either from the reads common to all the isoforms of a gene or by summing the isoform-level estimates derived from reads unique to individual isoforms