MMLmiRLocNet: miRNA Subcellular Localization Prediction Based on Multi-View Multi-Label Learning for Drug Design

Area of Science:

• Computational biology
• Molecular biology
• Bioinformatics

Background:

• Accurate identification of microRNA (miRNA) subcellular localization is crucial for understanding cellular mechanisms and developing targeted therapies.
• Existing computational methods often rely on a single data perspective, limiting their ability to integrate heterogeneous network information.

Purpose of the Study:

• To develop a novel computational method, MMLmiRLocNet, for enhanced prediction of miRNA subcellular localization.
• To address the challenge of fusing multi-view, heterogeneous network data in miRNA localization prediction.

Main Methods:

• MMLmiRLocNet employs a multi-view, multi-label learning strategy.
• It extracts sequence representations from lexical (k-mer physicochemical profiles), syntactic (word2vec embeddings), and semantic (pre-trained embeddings) aspects.
• A module for extracting consensus and specific features from multiple views was developed.

Main Results:

• MMLmiRLocNet demonstrated superior performance compared to existing methods, achieving high F1, subACC, and Accuracy scores.
• The integration of multi-view consensus and specific features significantly enhanced prediction accuracy.
• The method effectively captures both shared and unique characteristics across different data perspectives.

Conclusions:

• MMLmiRLocNet offers a robust and accurate computational approach for predicting miRNA subcellular localization.
• The multi-view feature extraction strategy is key to its improved performance.
• This advancement has significant implications for miRNA functional studies and therapeutic strategies.