Summary: Human alpha satellite and satellite 2/3 contribute to several percent of the human genome. However, identifying these sequences with traditional algorithms is computationally intensive. Here we develop dna-brnn, a recurrent neural network to learn the sequences of the two classes of centromeric repeats. It achieves high similarity to RepeatMasker and is times faster. Dna-brnn explores a novel application of deep learning and may accelerate the study of the evolution of the two repeat classes.
Availability and implementation: https://github.com/lh3/dna-nn.
© The Author(s) 2019. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.