Splice Junction Identification using Long Short-Term Memory Neural Networks.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Alternative splicing is a crucial process in generating mature messenger RNA from pre-messenger RNA in multi-exon genes.
• Identifying splice junctions is vital due to the high prevalence of alternative splicing in eukaryotic genomes.

Purpose of the Study:

• To develop a deep learning model for accurate identification of splice junctions in RNA sequences.
• To utilize a dataset of 13,666 unique primate RNA sequences for model training and validation.

Main Methods:

• A Long Short-Term Memory (LSTM) Neural Network was developed to classify RNA sequences.
• The model was trained on trinucleotide groups and evaluated using validation and test datasets to ensure robustness.
• Classification categories include Exon-Intron (EI), Intron-Exon (IE), and No splice (N).

Main Results:

• The finalized LSTM model demonstrated strong performance, achieving an average accuracy of 91.34%.
• The model also achieved an average f-score of 91.36% across 50 independent runs.
• Performance was rigorously assessed using accuracy and f-score metrics on test data.

Conclusions:

• The developed LSTM model is highly competitive with current Convolutional Neural Network (CNN) based approaches for splice junction identification.
• This study presents the LSTM model with the highest reported accuracy and f-score among similar LSTM-based structures in the literature.
• The findings highlight the potential of deep learning, specifically LSTMs, for advancing splice site prediction in RNA sequences.