Fig. 2.

Identification of structural errors in protein-coding gene predictions. A screen capture of the Apollo genome annotation editor showing a set of manually curated genes (located on chromosome X from 14,314,000 to 14,325,000) and their underlying evidence. Four individual tracks are displayed from top to bottom: BRAKER gene models, StringTie gene models, PacBio Iso-Seq refined transcript alignments, and paired-end Illumina RNA-seq alignments. The final set of curated gene models is displayed in the top box shaded in yellow labeled ‘User-created Annotations’. Both Illumina and PacBio RNA data suggest that the two BRAKER genes at the ends of this region were incorrectly split. StringTie models resolve the incorrect split but lack the two internal genes on the opposite strand (g2618.t1 and g2619.t1), because they lack long-read RNA coverage. We kept the StringTie model that best matches the RNA evidence and added the two internal genes on the opposite strand predicted by BRAKER and supported by short-read RNA-seq. Curated and predicted gene models are colored by coding sequence phase. Illumina and IsoSeq alignments are colored by strand orientation