Predicting MicroRNA Sequence Using CNN and LSTM Stacked in Seq2Seq Architecture.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Convolutional Neural Networks (CNN) excel at feature extraction.
• Long Short-Term Memory (LSTM) networks are adept at natural language processing.
• RNA sequences, including messenger RNA (mRNA) and microRNA (miRNA), can be analyzed using computational methods.

Purpose of the Study:

• To develop a computational model for predicting microRNA (miRNA) sequences based on messenger RNA (mRNA) sequences.
• To leverage the feature extraction capabilities of CNNs and the sequence processing power of LSTMs for RNA sequence analysis.
• To identify key sequence features, such as seed matches and G-U wobble base pairs, involved in miRNA-mRNA interactions.

Main Methods:

• Utilizing a CNN for feature extraction from mRNA sequences.
• Employing an LSTM network to predict miRNA sequences based on extracted mRNA features.
• Training and validating the model on experimentally validated miRNA-mRNA interaction data.
• Analyzing predicted target gene expression changes using microarray data.

Main Results:

• The model successfully learned to identify essential features, including seed matches (nucleotides 2-8) and G-U wobble base pairs in the seed region.
• The predictive model achieved an average accuracy of 72% in identifying miRNAs for specific mRNA sequences.
• The predicted miRNA targets exhibited the highest positive expression fold change compared to other prediction tools on microarray data.

Conclusions:

• The combined CNN-LSTM model demonstrates significant potential for accurate miRNA sequence prediction from mRNA sequences.
• The model's ability to identify critical sequence features highlights its biological relevance.
• This approach offers a promising tool for advancing research in gene regulation and RNA biology.