Three Methods to Transfer Annotation between Species or Strains

The accuracy with which gene structures in new target genomes can be predicted can be improved when leveraging a high quality annotation of a related genome. In the talk I will present and compare three conceptually different methods for this task. The first method, TransMap, uses a pairwise genome alignment to map transcript sequences, including alternative splice variants, from a source genome to the target genome, and is followed by small corrections only, optionally supported by target specific RNA-Seq. Another method - comparative AUGUSTUS - uses a multiple genome alignment, a phylogenetic tree, existing annotations for a subset of the genomes, RNA-Seq for another subset of the genomes as well as ab initio model components to identify gene structures in all genomes. The gene structures are preferred to agree on aligned exon boundaries. A third method uses spliced alignments of protein sequences, e.g. from GenomeThreader or exonerate, to produce evidence for conserved parts of a gene structure, which is then jointly considered with other evidence by AUGUSTUS. The three approaches aim at different degrees of similarity between the target genome and the source genome(s) and proteins, respectively. We applied the first two methods to 18 de novo assembled mouse strains and tested the last two methods on the Drosophila clade.