PREDICTION OF ALTERNATIVE SPLICING EVENTS ON PANTRANSCRIPTS

Abstract

Alternative Splicing (AS) is a regulation mechanism that contributes to protein diversity and is also associated to many diseases and tumors. Alternative splicing events quantification from RNA-Seq reads is a crucial step in understanding this complex biological mechanism. However, tools for AS events detection and quantification show inconsistent results. This reduces their reliability in fully capturing and explaining alternative splicing. Pangenomes have revolutionised bioinformatics by accommodating genetic diversity within populations, reducing reliance on single reference genomes. Extending this concept to the transcriptome, this study introduced an innovative Approach for predicting alternative splicing (AS) events, leveraging graph theory to map RNA-Seq data onto a pan-transcriptome graph. The objectives were twofold: first, to create an innovative method for AS event detection and prediction, and second, to assess its performance against simulated and real data, as well as compare it to the state-of-the art rMATS tool. The study constructed a specialised transcriptome graph using the vg tool and aligned RNA-Seq reads directly to it, obviating the need for read assembly. Differential quantification of AS events was conducted using specialised tools and benchmarked against established methods on simulated and real RNA-Seq datasets. The approach was initially tested on Drosophila data, a widely-used model organism, and subsequently validated using real sequencing data from Homo sapiens. The Approach performed competitively with rMATS in precision and recall across different event types, achieving precision values ranging from 0.624 to 0.941 and recall values ranging from 0.771 to 0.958. For exon skipping events, the Approach demonstrated higher precision (0.941) compared to rMATS (0.964), while maintaining a comparable recall of 0.958. Overall, the Approach showed commendable accuracy in detecting both annotated and novel alternative splicing events, making it a robust tool for transcriptome analysis with precision rates exceeding 96% for most event types. The Approach represents a significant advancement in AS event prediction, offering a versatile pipeline capable of handling complex transcriptome graphs. Its effectiveness in identifying diverse AS events underscores its potential to deepen the understanding of RNA splicing mechanisms and genetic diversity within populations. By providing a refined view of AS events within population-based transcriptomes, it offers a promising platform for future bioinformatics and transcriptome research endeavours.