Gene Finding Platform for Large Plant and Animal Genomes

In this talk we describe a new platform for automatic gene finding in large genomes of plants and animals. The input includes available genomic and transcriptomic data. The list of protein-coding genes (exon-intron structures, transcript isoforms) is created as an output.

The gene finding platform uses a combination of established tools to solve some common pre-processing tasks, such as de novo repeat masking, transcriptome to DNA alignment, etc. Data on reconstructed transcriptome, filtered and annotated, are used for genomic sequence labeling. Depending on the amount of the labeled sequence, one of the three methods of estimation of species specific parameters for gene prediction is selected: unsupervised, semi-supervised or supervised training. Parameter estimation is performed in one step or multiple iterations. In the final step of gene annotation, the transcriptome data is used again for profiling ab initio gene prediction in regions with reliable transcription activity.

The core part of the platform is adaptation of the training procedure to variable amounts and quality of input data. The goal is to make the platform universally useful both at the initial stages of a sequencing project, with little or no transcriptome data available, as well as at the stages when a complete genome sequence is available along with the large body of transcriptome data, or at any stage in between.