misoTar: a novel approach for predicting miRNA and isomiR targets.

Understanding microRNA/isomiR-mRNA interactions has long been a major challenge. Although many computational methods exist for predicting miRNA-mRNA interactions, almost all fail to account for isomiR-mRNA interactions. To bridge this gap, we developed misoTar, a fine-tuned BERT-based deep learning model trained on over 6.660 million positive and negative microRNA/isomiR-mRNA interaction pairs from 67 publicly available human samples across six studies. In five-fold cross-validation, misoTar achieved an average precision of 0.930 and a recall of 0.898. On independent test datasets, it consistently delivered superior or comparable performance relative to existing tools, including TargetScan, Mimosa, DMISO, and TEC-miTarget. Furthermore, single-nucleotide mutation analysis of true positive interactions highlighted the critical functional importance of non-seed regions in microRNA/isomiR-mRNA targeting. Overall, misoTar offers a robust and accurate framework for predicting microRNA/isomiR-mRNA interactions and provides new insights into microRNA biology. The misoTar tool is available at https://figshare.com/projects/misoTar/262723.