New Paper: DeepGRP: engineering a software tool for predicting genomic repetitive elements using Recurrent Neural Networks with attention
23 August 2021
Repetitive elements contribute up to 50% of eukaryotic genomes. So identifying and classifying repeats is an important step in genome annotation. In collaboration with Fabian Hausmann (Inst. for Medical Systems Biology, University Medical Center Hamburg-Eppendorf) the Genome Informatics group published a paper on a new software tool DeepGRP to annotate repetitive elements. It combines recurrent neural networks with techniques developed for neural machine translation. An evaluation on the human genome shows that DeepGRP achieves 20% improvement in prediction quality and improved running times, compared to previous similar methods. It is in particular able to transfer annotations learned from human data to the mouse genome. For details see https://doi.org/10.1186/s13015-021-00199-0 Hausmann, F., Kurtz, S. DeepGRP: engineering a software tool for predicting genomic repetitive elements using Recurrent Neural Networks with attention. Algorithms Mol Biol 16, 20 (2021)