Biocuration Improves Annotation of Protein-Coding Genes in the Sugarcane Genome

The sequencing of 319 BAC inserts of sugarcane R570 cultivar genome generated the prediction of 1,500 protein-coding genes not associated with mobile genetic elements. These were accomplished using a combination of softwares (Augustus; GlimmerHMM; PASA; Evidence Modeler; SignalP, TMHMM), databases (Phytozome; UniProt, InterPro), transcriptome data (SUCEST Project) and manual annotation of genes and transposable elements (TE).  The average number of genes defined for this set of 319 sequences of around 100 kbp was five, varying from none to eighteen. Exon-intron boundaries of each gene were manually inspected.  Transposable elements were annotated in most of the BACs in between genes and inside of intron sequences, found either as a complete version or as a partial element. Using this curated dataset a routine was established to annotate newly sequenced BAC inserts. RNA-seq data obtained from buds developing at five days after planting was also added to the analysis. The improved method included Exonarate, GMAP and LTR_STRUC predictions. Putative protein-conding genes were screened for false positives by BlastP analysis against UniProt and IntePro. Genes showing low or no similarities were automatically removed. Lastly, sequences were screened for false negative by BastX. The final prediction in 30 new sequences of the same cultivar counted from 1 to 16 genes. 34% of the total predictions were removed and 32% of all sequences were identified as TEs. This new set of predictions will be again curated and introduced into the routine to create an efficient semi-automatic annotation method to sugarcane genome annotation.