UniSplicer: a deep learning framework for accurate splice-site prediction and splice-altering mutation detection across diverse taxa

Conghao Hong, Wenzhen Cheng, Zhengyi Li, Jiajie Deng, Yiqiong Li, Youyi Zang, Hongbo Gao,
UniSplicer: a deep learning framework for accurate splice-site prediction and splice-altering mutation detection across diverse taxa,
Plant Communications,
2025,
101686,
ISSN 2590-3462,
https://doi.org/10.1016/j.xplc.2025.101686.
(https://www.sciencedirect.com/science/article/pii/S2590346225004481)
Abstract: ABSTRACT
RNA splicing removes non-coding introns from pre-mRNA to produce mature mRNA in eukaryotes. Accurate identification of splice sites is essential for the understanding of gene structures. Previous gene annotation and prediction heavily rely on the availability of high-quality genome assemblies, intensive functional studies and massive amount of resources, which restrict the analysis and application of the genomic sequences in various non-model species. Here, we present a deep learning-based model training framework that is able to develop accurate intron splice site prediction models for diverse species with relatively limited transcriptomic data. The UniSplicer-based models (http://www.unisplicer.com) outperform existing prediction models in various species, from plants to fungi and metazoans. UniSplicer-based models prediction scores could serve as reliable indicators of the effects of mutations in various types of splice mutants. Moreover, UniSplicer A. thaliana model identified genes in Arabidopsis ecotypes that exhibit abnormal splicing due to sequence variations near splice sites, which may be under environmental selection. Overall, UniSplicer-based models achieved high prediction accuracy and provided insights into how sequence variations result in splicing alteration of genes in large genomic data sets.
Keywords: RNA splice site prediction; deep learning; genome annotation; splicing alteration; cryptic splice site; environmental adaptation