MCTASmRNA: A deep learning framework for alternative splicing events classification.

Area of Science:

• Genomics and Molecular Biology
• Bioinformatics and Computational Biology

Background:

• Alternative splicing (AS) is a key post-transcriptional gene regulation mechanism in eukaryotes.
• Current AS detection methods are often inefficient, time-consuming, and struggle with RNA sequence complexity.

Purpose of the Study:

• To develop an efficient and accurate model for classifying alternative splicing events in mRNA sequences.
• To overcome limitations of existing AS detection tools, particularly in handling complex RNA sequences and large datasets.

Main Methods:

• Evaluated 10 AS detection tools, selecting rMATS for dataset construction.
• Developed a multi-scale convolutional and Transformer-based model (MCTASmRNA) for AS event classification.
• Incorporated an efficient channel attention mechanism and a novel joint loss function for model optimization.

Main Results:

• MCTASmRNA achieved superior performance compared to baseline models, demonstrating significant accuracy improvements.
• The model exhibited enhanced generalizability across different species, proving effective without a reference genome.
• Successfully addressed challenges of large intra-class and small inter-class differences in AS event sequences.

Conclusions:

• MCTASmRNA offers a powerful and efficient approach for alternative splicing event classification.
• The model provides valuable support for AS research across diverse organisms, enhancing our understanding of gene regulation.
• Future work will focus on model optimization and expansion for deeper exploration of AS mechanisms.