Decoding exonic intron retention from sequence: an exploratory study in the human genome

Alternative splicing enables the production of multiple transcripts from a single gene. Among its major forms, intron retention is particularly well characterised in plants, but with ample evidence also in animals. While most introns are constitutively spliced to ensure efficient removal from protein-coding gene transcripts, a subset exhibits an intrinsic sequence-level predisposition to be selectively retained under specific cellular or environmental conditions. However, the synergistic interplay of sequence features that creates this intrinsic predisposition remains only partially understood. Here, we present an exploratory, sequence-centric analysis of intron retention in the human genome. We systematically extract exon-intron-exon units from high-confidence transcript annotations and characterize each event using a broad set of sequence-derived features, including intron and exon architecture, splice site strength, nucleotide composition, transposable element content, and RNA-binding protein (RBP) binding patterns. Using integrative Random Forest and LASSO frameworks, we evaluate both the individual and joint discriminative power of these features. Our results show that intrinsic predisposition to intron retention cannot be explained by any single determinant, but instead emerges from the synergistic interplay of multiple factors. Intron length and GC content represent dominant predictors, while splice site strength, specific transposable element families, and structured RBP binding signatures provide additional explanatory power. Notably, RBP-derived features alone retain substantial predictive capacity, highlighting the importance of regulatory signal organization beyond basic sequence composition. Overall, this study provides a unified view of the sequence-level determinants underlying intron retention in humans and establishes a scalable computational framework for a deeper understanding of post-transcriptional gene regulation and its potential dysregulation.