Complexity of Gene Landscape Beyond Model Organisms

The view most of us have about genes models, from working on model organisms, ends up rather simplistic, with genes following one another nicely, leaving space for regulatory regions, mostly spread over large portions of the whole genome interspersed with repeats. Isochores are found on the human genome, with regions showing different GC% which affect gene models to some extent.  But from work on a larger diversity of organism, spanning from plants to animals, mostly not (yet) model organisms we see a great deal of complexity. With genes organized unevenly across the genome, from very compact parts of the genome, leaving no space between coding genes to parts of the genome scattered with distant exons covering 100s of Kb. Other genomes, relatively small genomes, use the ‘empty space’ from introns to code for other genes, leaving intricate gene models. These findings are the result from the much broader insight RNAseq offers relative to transcription, and the fact that reads didn’t match gene-models although mapped on the genome. Some classified part of these these as ‘leaky transcription’, but when looking closer these end up to be missed annotations due to inappropriate gene callers. All these new gene models and organization of the genome challenges the linear understanding of genes and their regulation, and thus also the gene prediction itself. Current gene-callers, using GHMMs and schemes, were designed based on a limited number of genomes studied. These gene-callers lack e.g. possibilities to start a new gene-model while the current one is not yet finished, leading to either broken genes or missing genes. The talk will give an overview of unexpected complexity we encountered while polishing annotations, hoping to fire interest of people developing gene-callers, and for users to be alert enough when using annotations.