P0377Simultaneous Gene Finding in Aligned Genomes
Mario Stanke
,
University of Greifswald, Greifswald, Germany
Stefanie König
,
University of Greifswald, Greifswald, Germany
Lizzy Gerischer
,
University of Greifswald, Greifswald, Germany
Lars Romoth
,
University of Greifswald, Greifswald, Germany
In many recent sequencing projects entire clades are sequenced whose member genomes are close enough to have widely conserved gene structures, yet diverse enough so information from mutation and selection can be exploited. We have extended the gene-finder AUGUSTUS for comparative gene prediction. The new AUGUSTUS-CGP takes a whole-genome alignment of multiple closely related species and simultaneously predicts the protein-coding genes in
all input genomes. This approach is conceptually different from previous programs which model only the structure of a single (N-SCAN, CONTRAST) or two genomes (pair-HMMs) at a time. Joint gene structures in all input genomes are scored considering the phylogeny, conservation of sequence and exon boundaries and selective pressure. Furthermore, evidence from content and signal models and evidence from RNA-Seq or existing annotations are considered where available.
We applied AUGUSTUS-CGP to predict the genes in 12 Drosophila genomes aligned with Cactus and evaluated
it on D.melanogaster. The comparative approach improves the accuracy of AUGUSTUS single-species gene finding both when no extrinsic evidence is used (ab initio) and when RNA-Seq data is included for a subset of the input genomes. In the de novo category, where only the aligned genomes are input, AUGUSTUS-CGP outperforms N-SCAN. The task of exploiting a trusted annotation of a related species for the annotation of a new target genome, can be considered a special case of comparative gene-finding. AUGUSTUS-CGP outperforms the homology-based gene-finder GeneWise for transferring the annotation of D.simulans to D.melanogaster.