MkcDBGAS: a reference-free approach to identify comprehensive alternative splicing events in a transcriptome

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Alternative splicing (AS) is a crucial post-transcriptional regulatory mechanism influencing diverse biological processes.
• Identifying all AS event types comprehensively, especially without a reference genome, remains a significant challenge in transcriptomics.

Purpose of the Study:

• To develop a novel, accurate, and scalable method for identifying all seven types of alternative splicing events using only transcriptome data.
• To overcome the limitations of existing tools that often require a reference genome.

Main Methods:

• Proposed MkcDBGAS, a three-module method utilizing a colored de Bruijn graph with dynamic- and mixed- kmers for AS event detection.
• Employed exon motifs and an XGBoost classifier for accurate AS type classification, achieving >93.40% accuracy.
• Validated the method's scalability and performance on Iso-Seq and transcriptome data from various species.

Main Results:

• MkcDBGAS precisely identifies AS events with >98.17% precision, detecting types missed by other tools.
• The XGBoost classifier accurately categorizes AS types, outperforming existing machine learning models and state-of-the-art methods.
• Demonstrated high scalability and effectiveness across diverse datasets, including human, Arabidopsis thaliana, Amborella, and mouse.

Conclusions:

• MkcDBGAS is the first accurate and scalable method capable of detecting all seven AS event types solely from transcriptome data.
• This novel approach significantly enhances the study of alternative splicing across a broader range of biological research fields.