MicroRNAs (miRNAs), vital regulators of gene expression and human health, are intimately associated with diseases upon dysregulation. Small molecules have emerged as promising therapeutic agents capable of modulating ...
详细信息
ISBN:
(纸本)9798400713026
MicroRNAs (miRNAs), vital regulators of gene expression and human health, are intimately associated with diseases upon dysregulation. Small molecules have emerged as promising therapeutic agents capable of modulating miRNA expression. However, laboratory-based identification of these regulatory effects is costly and time-consuming, highlighting the need for efficient and accurate computational models. Therefore, we propose DeepDrugmiR, a two-stage deep learning predictor that revolutionizes the prediction of small molecule-miRNA associations and their regulatory effects on miRNA expression. Leveraging novel language models tailored for chemical and DNA data, DeepDrugmiR encodes small molecules using double chemical language models and miRNAs using DNABERT-2, a nucleic acid language model. Its unique application of capsule networks, a next-generation neural architecture, further enhances prediction accuracy by capturing intricate relationships within small molecule structures and miRNA sequences. On benchmark datasets, DeepDrugmiR achieves state-of-the-art accuracy: 97% in identifying small molecule-miRNA associations and 92.7% (upregulation) & 92.4% (down-regulation) in predicting regulatory effects. This two-stage framework surpasses traditional single-stage methods in both information processing efficiency and noise resilience. A sorafenib case study validates DeepDrugmiR's usability, demonstrating consistency with miRNA profiling data. For ease of access, we have developed an intuitive web portal (https://***/similar to DeepDrugmiR/) to disseminate this innovative tool. By accelerating the understanding of drug-induced miRNA regulatory mechanisms, DeepDrugmiR propels the field of miRNA-targeted therapy forward.
暂无评论