Panthera: a deep learning pan-genomic splice haplotypes identification tool
- 1Institute of Molecular and Cell Biology, Agency for Science, Technology and Research, Singapore, 138673 Singapore
- 2Centre for Genomic Regulation, 08003 Barcelona, Spain
Abstract
Individual genetic polymorphisms can exert a modifier effect on splicing. With about one single-nucleotide polymorphism per 250 bases, we hypothesized that consideration of splice haplotypes will improve the accuracy of spliceogenic variant annotation, as current methods mainly assess single DNA variants and in a single reference genome. We describe Panthera for high-throughput pangenomic analysis of splice haplotypes, available as an open-source on GitHub (https://github.com/CherWeiYuan/Panthera). It accepts user-input variants and uses a deep learning model to predict their combinatorial effect on splice site probabilities under 64 non-redundant genetic backgrounds derived from five human super populations. The 874,587-parameter model is designed based on splicing mechanisms where multiple functional motifs, spliceosomal and auxiliary elements, interact with the sequential implementations of convolutional neural network (CNN) and Transformer blocks to simulate local and full-sequence motif interactions respectively. Panthera was validated to accurately detect every experimentally validated splice haplotype reported for FAS exon 6 and CFTR exon 10. Subsequently, two genes were used as case studies to demonstrate Panthera’s capability to identify novel splice haplotypes and novel splice modifiers therein. In MLH1, a haploinsufficient tumour suppressor whose loss of functional expression causes Lynch syndrome and colorectal cancers, Panthera identified a novel splice modifier in tandem with each of the two novel splice variants predicted to activate two respective novel frameshifting pseudoexons in an African individual. For the pathogenic PCCB c.654 + 462A > G splice variant implicated in propionic acidemia, Panthera identified a novel splice haplotype with a novel splice modifier, a CTGATGT insertion, that significantly enhances the pseudoexon activation in 55 genetic backgrounds. Notably, all predictions were experimentally validated through splicing assays on minigenes constructed with the respective splice haplotypes expressed in two cell lines. The case studies of FAS, CFTR, MLH1, and PCCB suggest haplotype differences can influence splicing, and a pangenomic haplotype tool is essential to detect them.
Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1038/s41598-026-65100-8.
Related
- GLASS: a graph learning algorithm for screening splice-aware alignments of long-read RNA-seq data
- A comparative analysis of graph neural networks and classical machine learning for carbapenem-resistance prediction and genomic interpretation in Acinetobacter baumannii
- The m6A motif is an anti-3' splice site
- Decoy and antisense RNA oligonucleotides targeting the EGFR exon 3 splice site suppress proliferation of A172 glioblastoma cells
- DNA augmentations for semi-supervised learning in genomic sequence classification.
BibTeX
@article{cher2026panthera,
title = {{Panthera: a deep learning pan-genomic splice haplotypes identification tool}},
author = {Wei Yuan Cher and Taniya Agarwal and Priscila Sun and Jin Rong Ow and Tommaso Tabaglio and Keng Boon Wee},
journal = {Scientific Reports},
year = {2026},
month = aug,
doi = {10.1038/s41598-026-65100-8},
url = {https://doi.org/10.1038/s41598-026-65100-8},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Panthera: a deep learning pan-genomic splice haplotypes identification tool
AU - Wei Yuan Cher
AU - Taniya Agarwal
AU - Priscila Sun
AU - Jin Rong Ow
AU - Tommaso Tabaglio
AU - Keng Boon Wee
PY - 2026
DA - 2026/08/14
JO - Scientific Reports
DO - 10.1038/s41598-026-65100-8
UR - https://doi.org/10.1038/s41598-026-65100-8
AB - Individual genetic polymorphisms can exert a modifier effect on splicing. With about one single-nucleotide polymorphism per 250 bases, we hypothesized that consideration of splice haplotypes will improve the accuracy of spliceogenic variant annotation, as current methods mainly assess single DNA variants and in a single reference genome. We describe Panthera for high-throughput pangenomic analysis of splice haplotypes, available as an open-source on GitHub (https://github.com/CherWeiYuan/Panthera). It accepts user-input variants and uses a deep learning model to predict their combinatorial effect on splice site probabilities under 64 non-redundant genetic backgrounds derived from five human super populations. The 874,587-parameter model is designed based on splicing mechanisms where multiple functional motifs, spliceosomal and auxiliary elements, interact with the sequential implementations of convolutional neural network (CNN) and Transformer blocks to simulate local and full-sequence motif interactions respectively. Panthera was validated to accurately detect every experimentally validated splice haplotype reported for FAS exon 6 and CFTR exon 10. Subsequently, two genes were used as case studies to demonstrate Panthera’s capability to identify novel splice haplotypes and novel splice modifiers therein. In MLH1, a haploinsufficient tumour suppressor whose loss of functional expression causes Lynch syndrome and colorectal cancers, Panthera identified a novel splice modifier in tandem with each of the two novel splice variants predicted to activate two respective novel frameshifting pseudoexons in an African individual. For the pathogenic PCCB c.654 + 462A > G splice variant implicated in propionic acidemia, Panthera identified a novel splice haplotype with a novel splice modifier, a CTGATGT insertion, that significantly enhances the pseudoexon activation in 55 genetic backgrounds. Notably, all predictions were experimentally validated through splicing assays on minigenes constructed with the respective splice haplotypes expressed in two cell lines. The case studies of FAS, CFTR, MLH1, and PCCB suggest haplotype differences can influence splicing, and a pangenomic haplotype tool is essential to detect them. Supplementary Information: The online version contains supplementary material available at https://doi.org/10.1038/s41598-026-65100-8.
SN - 2045-2322
ER -