MirLocPredictor: A ConvNet-Based Multi-Label MicroRNA Subcellular Localization Predictor by Incorporating k-Mer...

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• MicroRNAs (miRNAs) are crucial regulators of gene expression, impacting approximately 60% of mammalian genes.
• Current methods for visualizing miRNA subcellular localization are limited, hindering research into their function, transport, and biogenesis.
• Existing tools like MIRLocator use sequence-to-sequence models and k-mer embeddings, but overlook nucleotide positional importance.

Purpose of the Study:

• To address the limitations in miRNA subcellular localization prediction.
• To introduce a novel sequence representation that captures nucleotide positional information.
• To develop an accurate computational tool for predicting miRNA subcellular localization.

Main Methods:

• Proposed kmerPR2vec, a novel representation fusing k-mer positional information with randomly initialized neural embeddings.
• Developed MirLocPredictor, an end-to-end system combining kmerPR2vec with Convolutional Neural Networks (CNNs) for miRNA localization prediction.
• Evaluated kmerPR2vec using deep learning models (CNN, RNN) and nine evaluation metrics.

Main Results:

• The kmerPR2vec representation demonstrated richer semantic information and greater discriminative power compared to existing methods.
• MirLocPredictor significantly outperformed state-of-the-art methods in miRNA subcellular localization prediction.
• Achieved substantial improvements of 18% in precision and 19% in recall.

Conclusions:

• The kmerPR2vec approach effectively incorporates crucial positional information for RNA sequences.
• MirLocPredictor offers a powerful and accurate tool for predicting miRNA subcellular localization.
• This advancement facilitates deeper understanding of miRNA functions and regulatory mechanisms.