Helixer: ab initio prediction of primary eukaryotic gene models combining deep learning and a hidden Markov model
Felix Holst, Anthony Bolger, Felicitas Kindel, Christopher Günther, Janina Maß, Sebastian ST Triesch et al. · 2025 · 153 citationsRead the paper
The accurate identification of genes is vital for understanding biological function, yet this remains challenging across many newly sequenced or less-studied species. Here we present Helixer, an artificial intelligence-based tool for ab initio gene prediction that delivers highly accurate gene models across fungal, plant, vertebrate and invertebrate genomes. Unlike traditional methods, Helixer operates without requiring additional experimental data such as RNA sequencing, making it broadly applicable to diverse species. We show that Helixer's pretrained models achieve accuracy on par with or exceeding current tools, producing gene annotations that closely match expert-curated references across multiple evaluation metrics. Its design enables immediate use on genomes without retraining, providing an efficient, accessible solution for genome annotation in both research and applied settings. The tool is available as an open-source software for local installation via GitHub. An online web interface is also available as well as through the Galaxy ToolShed.
What the full text shows
Working prototype- Helixer outperforms AUGUSTUS in exon, gene, and intron F1 scores for a majority of tested species. — Exon F1 (9/16), Gene F1 (10/16), Intron F1 (8/16) species (Compared to AUGUSTUS with softmasking enabled)
- Helixer achieves higher recall than reference annotations for plant species, though lower precision. — Recall 0.958 vs Reference 0.931; Precision 0.878 vs Reference 0.966 (Plant species test set using Mapman4 protein annotations)
- Helixer is significantly faster than AUGUSTUS and GeneMark-ES in single-threaded mode. — 6.2–20.1 fold lower wall time (Smallest/largest fungus and smallest plant test genomes)
- Helixer identifies true annotations missing from established references in A. thaliana. — 102 true annotations identified vs 93 for Araport11 (Validated by conservation in at least 20 streptophyte species)
- Tiberius outperforms Helixer specifically in the Mammalia clade. — 20% higher gene recall and precision (Mammalian genomes (H. sapiens, B. taurus, D. leucas))
- Setup
- Trained and evaluated on 936 genomes across fungi, plants, vertebrates, and invertebrates using RefSeq and Phytozome13 data.
- How
- A sequence-to-label neural network (convolutional and recurrent layers) predicts base-wise genomic features, which are then processed by a hidden Markov model (HelixerPost) to assemble final gene models. The system is trained end-to-end on high-quality reference annotations without requiring RNA-seq or protein homology during inference.
- Limitations
- Lower precision in defining gene boundaries in fragmented genomes. · Lower precision compared to data-supported reference pipelines. · Less efficient multithreading compared to some existing tools.
Summarized by Gemma 4 from the open-access full text · 10 Oct 2026. Read it