Annotation of Protein Coding Genes for Non-Model de-novo Sequenced Organisms

Sequencing technologies have enabled genetic studies on a wide range of living organisms. Together with genome assemblies, accurate gene model predictions are crucial elements for helping answering all kinds of biological questions. For protein-coding gene (PCG) annotation, our group has been employing a semi-automatic pipeline that combines many sources of evidence in order to build consensus gene and transcript models. The complete annotation pipeline includes estimation of completeness of the assembly (i.e. “gene space”); masking of complex repeats; combination of gene prediction tools (including their training and evaluation), mapping of available transcripts and proteins, prediction of particular classes of proteins and assignment of possible functions. This pipeline uses broad range of tools. A number of these tools have been developed in-house. These programs include the ab-initiogene prediction program geneid, homology evidence-based tool SGP2, and the program Selenoprofiles. Selenoprofiles was specially designed for identification of selenoprotein genes, generally mispredicted by standard annotation pipelines due to the presence of an in-frame TGA codon, normally a stop, in their coding sequence.

We have applied our genome annotation approach on a wide variety of the de-novo sequenced organisms - vertebrates, plants, insects and fungus. Quality of the PCGs annotation is highly dependent on the quality of the available data. For example, if exhaustive transcript data is available and the degree of completeness/quality of the assembly is very high we have been able to obtain up to 96% of complete gene models with ~70% of these having annotated UTRs and ~25% alternative splicing forms. Therefore, we expect that the most comprehensive the transcriptional data the most we can improve the gene models themselves.