GeneMark-EP and -EP+: eukaryotic gene prediction with self-training in the space of genes and proteins

Abstract

<jats:p>We have made several steps towards creating a fast and accurate algorithm for gene prediction in eukaryotic genomes. First, we introduced an automated method for efficient <jats:italic>ab initio</jats:italic> gene finding, GeneMark-ES, with parameters trained in iterative <jats:italic>unsupervised</jats:italic> mode. Next, in GeneMark-ET we proposed a method of integration of unsupervised training with information on intron positions revealed by mapping short RNA reads.</jats:p><jats:p>Now we describe GeneMark-EP, a tool that utilizes another source of external information, a protein database, readily available prior to a start of a sequencing project. A new specialized pipeline, ProtHint, initiates massive protein mapping to genome and extracts hints to splice sites and translation start and stop sites of potential genes. GeneMark-EP uses the hints to improve estimation of model parameters as well as to adjust co-ordinates of predicted genes if they disagree with the most reliable hints (the -EP+ mode).</jats:p><jats:p>Tests of GeneMark-EP and -EP+ demonstrated improvements in gene prediction accuracy in comparison with GeneMark-ES, while the GeneMark-EP+ showed higher accuracy than GeneMark-ET. We have observed that the most pronounced improvements in gene prediction accuracy happened in large eukaryotic genomes.</jats:p>

BRAKER2: Automatic Eukaryotic Genome Annotation with GeneMark-EP+ and AUGUSTUS Supported by a Protein Database

“…First, the ab initio gene finder GeneMark-ES [1] completes self-training on a given genome and delivers predicted genes, the initial set of seed genes. This step is a part of the internal pipeline ProtHint (described earlier [26]) that executes GeneMark-ES, DIAMOND [36] and Spaln [31] (Fig. 1).…”

“…The connection between translated seed genes (seed proteins) and the genomic loci where the seed genes are residing (seed regions) is used in the subsequent steps. The seed proteins make queries for the DIAMOND database search that identifies potentially homologous (target) proteins in a protein database [26]. BRAKER2 runs in two major iterations (Fig.…”

“…proteins. The accuracy of a gene finding algorithm, which utilizes cross-species proteins mapping, depends strongly on the evolutionary distance between the species [10,26,45].…”

“…Nonetheless, leveraging of large volumes of proteins, particularly proteins from remotely related species, for improving genome annotation appears to be an important theoretical and practical task. Recently developed GeneMark-EP+ [26] was able to direct self-training by hints created by processing spliced alignments of large numbers of cross-species proteins. It was logical to integrate GeneMark-EP+ and AUGUSTUS in a pipeline relying on already known protein sequences as external evidence.…”

“…However, it turned out that the accuracy of the spliced alignment deteriorated quickly with increase of the evolutionary distance between two species. GeneMark-EP+ [26] has addressed this problem by the use of spliced alignments of multiple proteins originating from either closely or remotely related species.…”