MMFmiRLocEL: A Multi-Model Fusion and Ensemble Learning Approach for Identifying miRNA Subcellular Localization Using...

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• MicroRNA subcellular localizations (MSLs) are crucial for understanding microRNA functions.
• Existing computational methods for MSL prediction often rely solely on sequence data and single-model approaches, limiting accuracy.
• There is a need for methods that incorporate diverse biological information, including RNA 3D structure.

Purpose of the Study:

• To develop a novel deep learning-based computational method for accurate MSL identification.
• To integrate sequence, RNA 3D structure, and functional association data for enhanced prediction.
• To improve upon the performance of existing MSL prediction tools.

Main Methods:

• Developed MMFmiRLocEL, a deep learning approach utilizing multi-model fusion and ensemble learning.
• Incorporated RNA 3D structure information via predicted structural models.
• Employed convolutional neural networks for sequence-based prediction and deep residual networks for function-based prediction.
• Integrated predictions from sequence, structure, and function models using weighted ensemble strategies.

Main Results:

• MMFmiRLocEL demonstrated superior performance compared to existing state-of-the-art methods for MSL identification.
• Ablation analysis confirmed the effectiveness of the multi-model fusion mechanism in enhancing predictive accuracy.
• The integration of sequence, structure, and function data significantly contributed to improved prediction outcomes.

Conclusions:

• MMFmiRLocEL represents a significant advancement in computational MSL prediction by effectively integrating multi-modal biological data.
• The proposed method offers a more robust and accurate approach to understanding miRNA functions through their subcellular localization.
• Future research can build upon this multi-model fusion strategy for other RNA-related prediction tasks.