Fig. 6

Filtered two-pass alignment improves genome-guided annotation. a Scatterplot showing precision against recall for intron chains in genome-guided transcriptome annotations generated from alignments using StringTie2. Precision and recall scores were calculated against reference annotations filtered to include only transcripts for which at least one read was simulated. Reads were simulated from Arabidopsis (left) and human (right) nanopore DRS data aligned to the AtRTD2 and GRCh38 reference transcriptomes, respectively. b–e Stripplots with box and whiskers showing the number of correct transcripts assembled (left panels) and precision of transcripts assembled (right panels) for genome-guided transcriptome assembly using StringTie2. Two-pass alignment improved the precision and number of transcripts assembled for real nanopore DRS data for b Arabidopsis, c human, d mouse, and e yeast. For all boxplots, overlaid strip-plots are shown for individual samples. Each sample was assigned a unique marker so that the changes in each sample could be tracked between the one-pass, two-pass, and reference-guided alignments. Box and whiskers not shown for samples with less than 4 data points. Y limits vary between figures since within-figure (i.e., same species and sequencing technology) comparison is more important than between-figure comparisons