CL-PMI: A Precursor MicroRNA Identification Method Based on Convolutional and Long Short-Term Memory Networks

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• MicroRNAs (miRNAs) are key gene regulators, primarily acting as translational repressors in mammals.
• Identifying precursor microRNAs (pre-miRNAs) is challenging due to their short length and complex structures.
• Existing computational methods often neglect sequential and spatial characteristics, and suffer from dataset imbalance, limiting pre-miRNA identification accuracy.

Purpose of the Study:

• To develop an advanced computational method for accurate pre-miRNA identification.
• To address the limitations of existing methods, including feature extraction and dataset imbalance.
• To improve the understanding and identification of gene-regulating molecules.

Main Methods:

• Proposed a cascaded Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) framework named CL-PMI.
• Utilized CNN for automatic spatial feature extraction and LSTM for capturing sequential characteristics.
• Incorporated attention mechanisms for long-term dependence modeling and focal loss to mitigate dataset imbalance.

Main Results:

• The CL-PMI algorithm demonstrated superior performance across all tested datasets compared to existing methods.
• The method effectively identified pre-miRNAs by integrating both spatial and sequential information.
• Focal loss successfully addressed the challenge of imbalanced datasets in pre-miRNA identification.

Conclusions:

• CL-PMI offers a robust and effective approach for pre-miRNA identification.
• The integration of CNN and LSTM, along with attention mechanisms, enhances the model's ability to capture complex pre-miRNA features.
• This study provides a significant advancement in computational methods for identifying crucial gene-regulating molecules.