Extended Data Fig. 5. Assessments of transcript unit prediction and schematic illustration of PINTS.

a, The consistencies vary greatly between transcription units annotated in GENCODE (Annot.) and those predicted by different tools^58,59,85 (Pred.). Lines in the violin plot indicate the 25th, 50th, and 75th quartiles, respectively. b, Schematic plot of PINTS. i, Improvement of TSS identification resolution by focusing only on read ends and using zero-inflated Poisson (ZIP) models to fit local background to address the significantly increased sparsity of signals. The thin grey lines indicate sequencing reads with the 5′ ends highlighted in red. ii, The existence of other potential true peaks (pink) elevates the estimation of read density in the local background. iii, A schematic plot shows how IQR-ZIP works. The blue box shows the read density distribution of the local background; the purple dot shows the density of the peak to be tested; the pink dot shows the density of a potential true peak close to the peak to be tested, whose read density is a clear outlier and thus excluded from local background estimation.