Fig. 3

Comparison of accuracy across multiple methods for single-end RNA-seq. a A schematic showing the set of transcripts estimated by individual methods and the common set (not drawn to scale). b-e Four different evaluation criteria are applied on the common set of transcripts indicated in (a). IsoEM.mm0 and IsoEM.mm2 refer to IsoEM runs with up to 0 and 2 mismatches allowed, respectively. NEUMA.t1 and NEUMA.t50 refer to NEUMA runs with EUMA (length) cut-off 1 bp and 50 bp, respectively. eXpress.default and eXpress.nobias refer to eXpress runs with default and –no-bias option, respectively. (b) Pearson correlation coefficient (c) root mean squared error (d) number of false positives, i.e., isoforms with zero true expression and with 1 or larger value in log₂(σM_e + 1), where M_e is the estimated TPM (\( FPK{M}_i/{\displaystyle {\sum}_{i\in T}}FPK{M}_i\times {10}^6 \) for transcript i in the transcriptome T), (e) number of false negatives, i.e., isoforms with 1 or larger value in log₂(σM_t + 1), where M_t is the molecular fraction of the transcript × 10⁶, and zero estimated expression. (f) total number of transcripts whose expression level is reported. EMSAR, RSEM, Sailfish and IsoEM report all transcript levels. Error bars indicate standard deviation, not standard error of mean