ASPECT: Alternative Splicing Event Classification with Transformers.

Area of Science:

• Genomics and Molecular Biology
• Computational Biology
• Bioinformatics

Background:

• Alternative splicing (AS) generates diverse mRNA isoforms, crucial for proteomic diversity.
• Dysfunctional AS variants are linked to various diseases.
• Current AS research often overlooks specific event types and relies on limited computational approaches.

Purpose of the Study:

• To develop a computational framework for classifying alternative splicing events.
• To address limitations in existing methods for capturing regulatory context in splicing decisions.
• To improve the comprehensive characterization of AS and its role in human health and disease.

Main Methods:

• Developed ASPECT, an alternative splicing event classification framework.
• Utilized DNABERT-2 with Byte Pair Encoding (BPE) tokenization.
• Evaluated performance on binary alternative splicing event pair classification tasks.

Main Results:

• ASPECT demonstrated strong performance across classification tasks, measured by AUC, F1-score, and accuracy.
• The framework reliably discriminated between closely related splicing event types.
• ASPECT showed consistent performance on TCGA BRCA cancer-associated splicing events, validating its applicability.

Conclusions:

• ASPECT provides a robust method for alternative splicing event classification.
• The framework enhances the ability to study complex splicing events and their disease implications.
• This work contributes to a deeper understanding of alternative splicing in human health.