SpliceSelectNet: a hierarchical Transformer-based deep learning model for splice site prediction

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Accurate RNA splicing is vital for gene expression and protein function.
• Mutations causing aberrant splicing are linked to diseases like cancer.
• Current computational methods struggle with long-range dependencies in splice site prediction.

Purpose of the Study:

• To develop a deep learning model for accurate splice site prediction and aberrant splicing detection.
• To address limitations in handling long-range dependencies in existing computational tools.
• To create a biologically interpretable framework for understanding splicing regulation.

Main Methods:

• Developed SpliceSelectNet (SSNet), a hierarchical Transformer-based deep learning model.
• Integrated local and global attention mechanisms to capture proximal and distal regulatory signals.
• Utilized single-nucleotide resolution for precise splice site prediction up to 100 kb DNA sequences.

Main Results:

• SSNet achieved state-of-the-art performance in splice site prediction and aberrant splicing detection on benchmark datasets.
• In silico mutagenesis confirmed that SSNet's attention scores reflect functional sequence importance.
• Long-range sequence perturbation experiments demonstrated SSNet's ability to capture distal regulatory effects.

Conclusions:

• SSNet provides a biologically interpretable framework for modeling long-range splicing regulation.
• The model enhances the accuracy of splice site prediction and aberrant splicing detection.
• SSNet offers a powerful tool for understanding genetic disorders and developing therapeutic strategies.