Introduction
MiRNA subcellular localizations (MSLs) are essential for uncovering and understanding miRNA functions in various biological processes. In transcriptomic studies, miRNAs are typically localized in multiple subcellular locations. Despite significant advancements in RNA function prediction, structural information has not been fully incorporated into MSL prediction models. Moreover, traditional methods often rely on single-model approaches, which fail to capture the full complexity of biological systems, further hindering predictive accuracy and performance. In this study, we propose the MMFmiRLocEL model, which integrates multi-model fusion and ensemble learning techniques based on RNA sequence, structure, and functional prediction methods for the identification of MSL. First, we employed a multiple sequence representation approach to construct a sequence-based model using a multi-scale convolutional network. Second, we attempt to use predicted structural information to construct a structure-based model using a graph-attention network-based model. Finally, we constructed a representation of the RNA-disease association network and employed a residual model to build a function-based model. Moreover, we incorporate sequence-structure-functional models through a weighted averaging ensemble multi-output classifier to improve the robustness and accuracy of MSL identification. Experimental results show the performance of MMFmiRLocEL outperforms the other state-of-the-art methods, demonstrating its superiority over traditional single-model approaches.
