Optimized Cancer miRNA Biomarker Classification Using Dynamic Multi-Resolution Spatio-Temporal Attention Network (DMR-STANet) with Adaptive Transfer Learning

A cancer diagnosis and treatment approach needs biomarkers, and miRNA profiles are valuable diagnostic tools. This study introduces DMR-STANet to enhance miRNA biomarker categorization. A robust multi-resolution analysis and efficient spatio-temporal attention network ensure that DMR-STANet captures both low-level and holistic miRNA data characteristics. The network's Dynamic Multi-Resolution Feature Extraction enhances characteristics that may affect cancer classification by processing biomarker data in several resolutions. The proposed Adaptive Transfer Learning (ATL) architecture fine-tunes pre-trained layers to enable biomarker-specific pattern identification with minimum computational cost.The approach minimizes feature learning variability across time-series miRNA datasets using a Temporal Gradient Consistency (TGC) loss function to improve temporal consistency and resilience. A Semi-Supervised Fine-Tuning Module uses pseudo-labelled data to enhance training dataset sizes, a typical issue in biomedical investigations without labeled samples. The suggested DMR-STANet also outperforms previous approaches in experimental evaluations on cancer miRNA datasets with 96.8% classification accuracy. The model is noise-resistant and performs well with huge data, making it suitable for biomedical diagnosis.The suggested method may assist establish a new generation of high-accuracy biomarker classification solutions for miRNA in oncology, a tough task in cancer diagnosis. Using dynamic multi-resolution analysis and spatio-temporal attention. DMR-STANet advances computational biology and precision medicine with adaptive learning. This research suggests clever algorithms might change biomarker-based cancer diagnosis.