IsoGraph resolves coordinated transcript choice and gene abundance in the aging human brain.

Abstract

Relative transcript usage can change without a corresponding change in total gene expression, but resolving coordinated transcript choice across genes while separating it from abundance remains challenging. We present IsoGraph, a gene-centered framework that represents isoform usage and abundance as paired network channels, identifies gene modules, and quantifies their channel-specific contributions. In synthetic benchmarks, IsoGraph preferentially recovered network structure when coordinated transcript usage carried the planted signal, whereas abundance-based representations were more informative in abundance-dominated and RNA-degradation settings. Applied to BrainSEQ and GTEx adult brain transcriptomes, IsoGraph identified age-associated transcript-usage signals that persisted after conditioning on gene abundance and, in a region-dependent manner, after adjustment for estimated cellular composition. Modules were stable across donor subsets, and frozen module axes retained age associations across cohorts even when exact gene partitions differed. IsoGraph module context captured predictive information about held-out differential transcript usage beyond gene-wise evidence and conventional co-expression. Convergent evidence from structural annotation, short-read junction usage, independent long-read sequencing, and within-donor allelic analyses supported constituent transcript events. Disease colocalization further prioritized candidate isoform events at neurological and psychiatric disease loci, including an alternative terminal-exon event at PRDM2 associated with amyotrophic lateral sclerosis. Together, these results establish coordinated transcript usage as a reproducible component of transcriptomic organization and provide a framework for resolving its contribution alongside gene abundance and for tracing network-level signals to candidate transcript events.