Integration of Theorectical and Experimental Data on Gene Regulation and Gene Evolution for Improved Genome Annotation

Progress in NextGen sequencing particularily RNASeq has dramatically impacted quality and design of genome annotation pipelines - especially for protein-coding gene prediction. Current gene prediction pipelines seem to rely less on ab initio predictions, putting more weight on extrinisic evidence and usually predict copious transcript isoforms derived from genome-guided or de-novo transcript assemblies. In some cases, mapped transcript assemblies from deep-sequencing of the transcriptome are chosen as the sole source of gene models for a particular genome. While this clearly aims to increase the quality and validity of gene structures, it introduces problems in gene set completeness, locus definition and major/reference isoform definition.

Usually these issues should be resolved by combiner tools that are either part of gene prediction pipelines or are available as standalone tools. But the depth of current RNASeq and other evidence sources spikes the full complexity of gene regulation, with non-sense mediated decay (NMD), alternative transcription initiation, translated upstream open reading frames, sRNA-guided RNA cleavages etc into the complement of transcript evidences. In addition, depending of the sequencing depth and transcripts expression level, one transcript can be fragemented into several assemblies. A major aim of most genome annotations is to define a gene set with one reference isoform per locus for comparative genomics. In these cases e.g. choosing the longest transcript or protein might fail to select the major. Untackled, non-mRNA evidences blur the identity of the major isoform or even lead to gene fusions or truncation.

In my talk I like to present some of the strategies I have employed for defining the gene catalogue of the moss Physcomitrella patens and will show heuristics and machine learning approaches I've used for locus definition, model selection and NMD variant detection.