Gene body architecture predicts isoform diversity across eukaryotic genomes

Abstract

Alternative splicing generates transcript and isoform diversity, but the gene-level features that distinguish isoform-rich genes from simpler genes remain incompletely defined. We examined whether structural, regulatory, sequence-derived, and functional features jointly explain isoform complexity in the human genome and whether these relationships are conserved across mouse, Drosophila , and zebrafish genomes as well. Using curated transcript annotations and gene-level biomolecular features, we found that isoform diversity is pervasive across all four species but varies by genome and biotype. Across species, isoform-rich genes were consistently associated with longer introns, shorter exons, and increased exon-intron organization. UTR associations were more variable, with 3′ UTR length showing stronger positive associations in humans and Drosophila . In humans, isoform count was negatively associated with GC content and CpG island density and positively associated with protein-protein interaction network size. Machine learning based modeling predicted human isoform counts with strong performance and retained predictive signal in mouse. PCA and UMAP further showed that isoform-associated gene features form a continuous, structured manifold rather than discrete gene classes. These findings support a model in which isoform diversity is shaped by conserved gene-structural features and modulated by lineage- and biotype-specific regulatory architecture. Integrating predictive modeling, feature attribution, and dimensionality reduction provides complementary insight into the genomic organization underlying transcript diversity.