DeepHel: improving the accuracy and completeness of Helitron annotation with a multi-stage deep learning framework.

Authors

Yuyang Liu, Minghua Xu, Jianxin Wang

Published in

Bioinformatics (Oxford, England). Nov 10, 2026. Epub Nov 10, 2026.

Abstract

Accurate and comprehensive annotation of Helitrons remains challenging due to weak structural signals, ambiguous termini, sequence fragmentation, and complex nested insertions. Existing methods often exhibit limited sensitivity, high false-positive rates, and insufficient support for simultaneous detection of both HLE1 and HLE2, leading to incomplete and imprecise annotation.
We present DeepHel, a multi-stage deep learning framework that improves the accuracy and completeness of Helitron annotation by modeling intactness, boundary characteristic, and structural consistency. DeepHel formulates Helitron detection as a staged representation learning problem, integrating sequence features, structural signals, and multiple sequence alignment information across four coordinated modules: candidate identification, intact element recognition, boundary refinement, and Helitron classification, followed by nested insertion removal. Benchmarking across five eukaryotic genomes under three evaluation settings demonstrates that DeepHel achieves a superior balance between sensitivity and precision compared with existing tools, including HiTE, HELIANO, HelitronScanner, RepeatModeler2, and EDTA. In particular, DeepHel improves the recovery of intact Helitron families (completeness), while maintaining robust performance in challenging scenarios such as low-copy elements and genomes with complex nested insertions. Furthermore, DeepHel identifies previously unannotated Helitrons with well-defined structural features, providing additional insights into Helitron diversity and genome organization. These results show that DeepHel provides a robust and comprehensive solution for accurate and complete annotation of both HLE1 and HLE2 across diverse eukaryotic genomes.
DeepHel is freely available at Github: https://github.com/lyysxb/DeepHel, with the archived version at Zenodo: https://zenodo.org/records/22069775.

PMID:
43009184
Bibliographic data and abstract were imported from PubMed on 11 Nov 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.