DeepRibo : a neural network for precise gene annotation of prokaryotes by combining ribosome profiling signal and binding site patterns

Annotation of gene expression in prokaryotes of-ten finds itself corrected due to small variations ofthe annotated gene regions observed between differ-ent (sub)-species. It has become apparent that tradi-tional sequence alignment algorithms, used for thecuration of genomes, are not able to map the fullcomplexity of the genomic landscape. We presentDeepRibo, a novel neural network utilizing featuresextracted from ribosome profiling information andbinding site sequence patterns that shows to be aprecise tool for the delineation and annotation of ex-pressed genes in prokaryotes. The neural networkcombines recurrent memory cells and convolutionallayers, adapting the information gained from boththe high-throughput ribosome profiling data and ri-bosome binding translation initiation sequence re-gion into one model. DeepRibo is designed as a sin-gle model trained on a variety of ribosome profil-ing experiments, used for the identification of openreading frames in prokaryotes withoutaprioriknowl-edge of the translational landscape. Through exten-sive validation of the model trained on various setsof data, multiple species sequence similarity, massspectrometry and Edman degradation verified pro-teins, the effectiveness of DeepRibo is highlighted.

Publication year:2019

BOF-keylabel:yes

IOF-keylabel:yes

BOF-publication weight:6

CSS-citation score:1

Authors:National

Authors from:Higher Education

Accessibility:Open