LSTM

LSTM predicts microRNA (miRNA) sequences from messenger RNA (mRNA) sequences by combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks within a Sequence-to-Sequence (Seq2Seq) architecture to model mRNA→miRNA relationships.

Key Features:

• CNN feature extraction: Uses Convolutional Neural Networks to extract local sequence motifs and patterns from mRNA sequences.
• LSTM sequence prediction: Employs Long Short-Term Memory networks to model long-range dependencies and predict miRNA nucleotide sequences.
• Seq2Seq architecture: Integrates CNN and LSTM within a Sequence-to-Sequence framework to map input mRNA sequences to output miRNA sequences.
• Seed match detection: Captures seed matches in the initial 2-8 nucleotides at the 5' end of miRNAs.
• G-U wobble prediction: Predicts G-U wobble base pairs within the miRNA seed region.
• Performance metrics: Demonstrated an average prediction accuracy of 72% for specific mRNAs.
• Microarray validation: Showed the highest positive expression fold change among predicted targets when validated against microarray data generated using anti-25 miRNAs.

Scientific Applications:

• miRNA discovery: Predicts candidate miRNA sequences from mRNA inputs to support identification of regulatory small RNAs.
• Post-transcriptional regulation studies: Provides insights into miRNA-mediated regulation and post-transcriptional control of gene expression.
• Gene regulatory network analysis: Facilitates elucidation of miRNA–mRNA interactions within genomics and transcriptomics research.
• Experimental prioritization: Prioritizes potential miRNA candidates for experimental validation based on prediction outputs and microarray validation signals.

Methodology:

The method applies CNN-based feature extraction on mRNA sequences followed by LSTM-based sequence prediction within a Seq2Seq architecture.