DiCleavePlus: A Transformer-Based Model to Detect Human Dicer Cleavage Sites Within Cleavage Patterns

MicroRNAs (miRNAs) play a crucial role in posttranscriptional gene regulation. The biogenesis of mature miRNAs requires precise cleavage of precursor miRNAs (pre-miRNAs) by Dicer. Several computational approaches have been developed to predict human Dicer cleavage sites; however, important limitations persist. Cleavage Pattern-based models, which rely on short pre-miRNA subsequences, can only identify positive patterns in which the cleavage site is centrally located. Conversely, models that do not rely on Cleavage Patterns generally exhibit suboptimal performance. These limitations highlight the need for a more accurate predictor that fully exploits sequence and structural information from pre-miRNAs. In this study, we propose DiCleavePlus, a Cleavage Pattern-based framework for predicting Dicer cleavage sites on pre-miRNAs. DiCleavePlus takes an extended Cleavage Pattern together with the full-length pre-miRNA sequence from which it is derived as input. A Transformer-based encoder is employed to extract features from both the pattern and the pre-miRNA. Benchmarking experiments demonstrate that DiCleavePlus achieves accurate and robust performance in predicting human Dicer cleavage sites.