Deciphering ribosomal frameshifting determinants across species with a semi-supervised hybrid learning framework
4 affiliations
- 1SDU-ANU Joint Science College, Shandong University, Weihai, Shandong 264209, China
- 2Marine College, Shandong University, Weihai, Shandong 264209, China
- 3Institute of Evolution and Marine Biodiversity, Ocean University of China, Qingdao, Shandong 266003, China
- 4Department of Biology, Bradley University, Peoria 61625, USA
Abstract
Programmed ribosomal frameshifting (PRF) is a conserved translational recoding mechanism that expands proteome diversity and regulates important biological processes across viruses, prokaryotes, and eukaryotes. However, existing predictors are often limited by their reliance on extensive annotated data, poor performance in non-model organisms, and restricted applicability across species or frameshift types. Here, we present FScanpy, a hybrid machine-learning model integrating histogram-based gradient boosting (HistGB) and a bidirectional long short-term memory-convolutional neural network (BiLSTM-CNN) architecture for cross-species PRF prediction. The model integrates short-range codon-position features associated with ribosomal pausing and long-range sequence contexts that reflect structural constraints around candidate shift sites. Using semi-supervised learning on 8 049 candidate sequences, FScanpy reduces reliance on manual annotation and improves model generalization. Across multiple datasets, it achieved an area under the receiver operating characteristic curve (AUC) of 0.951, and it also performed well on euplotid protists (AUC=0.877), supporting its utility in non-model systems with non-canonical recoding patterns. Feature interpretation further showed that AAA and GGG motifs, together with mRNA secondary-structure stability, as major determinants across datasets, highlighting conserved features of PRF from viruses to eukaryotes. These insights not only validate the model’s accuracy but also deepen understanding of PRF’s conserved biological roles. FScanpy establishes a novel and effective computational framework for PRF analysis by resolving challenges in feature representation and species generalizability, enabling the study of frameshifting sites and accelerating discovery of evolutionarily divergent recoding events.
Related
- Targeting SARS-CoV-2 programmed −1 ribosomal frameshifting: structural dynamics and RNA-directed antiviral strategies
- METANet: a supervised ensemble learning framework for reconstructing direct and functional tissue-specific transcription factor networks.
- Structure-guided antisense-oligonucleotides selectively modulate frameshifting of a human gene
- AET5: A transcriptome-guided molecular generation framework with contrastive self-supervised learning
- Structural basis of ribosomal RNA repair by RtcB2
BibTeX
@article{yang2026deciphering,
title = {{Deciphering ribosomal frameshifting determinants across species with a semi-supervised hybrid learning framework}},
author = {Yu-Hao Yang and Juan Yang and Zi-Jia Liu and Yuan Li and Wei-Bo Song and Naomi A. Stover and Xiao Chen},
journal = {Zoological Research},
year = {2026},
month = sep,
doi = {10.24272/j.issn.2095-8137.2025.648},
url = {https://doi.org/10.24272/j.issn.2095-8137.2025.648},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - Deciphering ribosomal frameshifting determinants across species with a semi-supervised hybrid learning framework
AU - Yu-Hao Yang
AU - Juan Yang
AU - Zi-Jia Liu
AU - Yuan Li
AU - Wei-Bo Song
AU - Naomi A. Stover
AU - Xiao Chen
PY - 2026
DA - 2026/09/18
JO - Zoological Research
DO - 10.24272/j.issn.2095-8137.2025.648
UR - https://doi.org/10.24272/j.issn.2095-8137.2025.648
AB - Programmed ribosomal frameshifting (PRF) is a conserved translational recoding mechanism that expands proteome diversity and regulates important biological processes across viruses, prokaryotes, and eukaryotes. However, existing predictors are often limited by their reliance on extensive annotated data, poor performance in non-model organisms, and restricted applicability across species or frameshift types. Here, we present FScanpy, a hybrid machine-learning model integrating histogram-based gradient boosting (HistGB) and a bidirectional long short-term memory-convolutional neural network (BiLSTM-CNN) architecture for cross-species PRF prediction. The model integrates short-range codon-position features associated with ribosomal pausing and long-range sequence contexts that reflect structural constraints around candidate shift sites. Using semi-supervised learning on 8 049 candidate sequences, FScanpy reduces reliance on manual annotation and improves model generalization. Across multiple datasets, it achieved an area under the receiver operating characteristic curve (AUC) of 0.951, and it also performed well on euplotid protists (AUC=0.877), supporting its utility in non-model systems with non-canonical recoding patterns. Feature interpretation further showed that AAA and GGG motifs, together with mRNA secondary-structure stability, as major determinants across datasets, highlighting conserved features of PRF from viruses to eukaryotes. These insights not only validate the model’s accuracy but also deepen understanding of PRF’s conserved biological roles. FScanpy establishes a novel and effective computational framework for PRF analysis by resolving challenges in feature representation and species generalizability, enabling the study of frameshifting sites and accelerating discovery of evolutionarily divergent recoding events.
SN - 2095-8137
ER -