PMMS: Predicting essential miRNAs based on multi-head self-attention mechanism and sequences

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• MicroRNAs (miRNAs) are crucial regulators of biological processes, and identifying essential miRNAs is vital for understanding disease etiology and mechanisms.
• Traditional experimental methods for identifying essential miRNAs are costly and time-consuming, necessitating the development of efficient computational approaches.

Purpose of the Study:

• To develop and validate a novel computational method, PMMS, for predicting essential miRNAs.
• To leverage multi-head self-attention mechanisms and sequence-based features for enhanced miRNA identification.

Main Methods:

• PMMS integrates statistical and structural features to extract static miRNA features.
• Bi-directional Long Short-Term Memory (BiLSTM) networks and multi-head self-attention are employed to extract deep learning-based features from pre-miRNA sequences.
• A weighted attention mechanism refines deep learning features, which are then concatenated with static features for prediction using a multilayer perceptron model.

Main Results:

• The PMMS method achieved high prediction performance in five-fold cross-validation, with an Area Under the ROC Curve (AUC) of 0.9556, an F1-score of 0.9030, and an accuracy (ACC) of 0.9097.
• PMMS demonstrated superior performance compared to other existing computational methods for essential miRNA identification.

Conclusions:

• PMMS is a highly effective and accurate computational tool for identifying essential miRNAs.
• The developed method offers a promising alternative to traditional experimental approaches, accelerating miRNA research and disease mechanism studies.