Beyond Gene Expression: Introducing IsoGraph, a New Framework for Understanding Transcript Choice in the Aging Human Brain
We’re excited to share our new preprint, “IsoGraph resolves coordinated transcript choice and gene abundance in the aging human brain,” now available on bioRxiv!
In this work, we introduce IsoGraph, a computational framework designed to investigate an important but often overlooked dimension of transcriptomic biology: how genes coordinate their choice of RNA transcripts, as distinct from changes in overall gene expression.
Using human brain transcriptomic data, we demonstrate how IsoGraph can identify coordinated transcript-usage patterns associated with aging, evaluate their reproducibility, and connect these patterns to candidate molecular events relevant to neurological and psychiatric disease.
This is the HEART-GeN Lab’s second preprint. It was first-authored by Alexis Bennett with Elisa Kain Johnson, and Dr. Kynon J. M. Benjamin served as corresponding author.
Looking Beyond How Much a Gene Is Expressed
Traditional gene expression analyses typically focus on measuring changes in the total amount of RNA produced by individual genes. These approaches have transformed our understanding of molecular biology, but they capture only part of the picture.
Many genes produce multiple RNA transcripts, or isoforms, through processes such as alternative splicing and alternative transcript-end usage. These transcripts can differ in their structure and potential biological functions.
Importantly, a gene can change which transcripts it produces without substantially changing its overall expression level.
This distinction raises an important question: can coordinated changes in transcript choice reveal biological signals that conventional gene-level expression analyses miss?
Although existing methods can identify changes in transcript usage within individual genes, detecting coordinated transcript-choice programs across multiple genes, and distinguishing them from changes in total gene abundance, remains challenging. This is the problem IsoGraph was developed to address.
Making Transcript Networks Accessible
The ideal way to study coordinated transcript choice would be single-cell long-read sequencing, which captures full-length transcripts within individual cell types. That approach remains expensive and difficult to scale to large cohorts. The alternatives each give something up: standard single-cell short-read assays capture little full-length isoform information, and bulk long-read sequencing is limited in depth and throughput.
Bulk short-read RNA-seq, in contrast, already exists for large human cohorts such as GTEx and BrainSEQ. These data have real limitations, including ambiguous isoform quantification and no direct view of cell identity, but they are the data many research groups already have.
We built IsoGraph to work with these data. Any group with a bulk RNA-seq cohort can use it to ask network-level questions about transcript choice, and then follow up the prioritized events with higher-resolution assays.
Introducing IsoGraph: Connecting Transcript Choice and Gene Abundance
IsoGraph is a gene-centered computational framework that integrates two complementary dimensions of transcriptomic variation:
- Transcript usage: how the relative proportions of different RNA isoforms change within a gene.
- Gene abundance: how the overall expression level of that gene changes.
Rather than treating these signals as interchangeable, IsoGraph represents them as separate but connected channels within a multiplex network. This approach enables researchers to identify groups of genes, or modules, that exhibit coordinated molecular variation and determine whether those patterns are primarily driven by transcript choice, gene abundance, or a combination of both.
IsoGraph also provides tools for network inference, module interpretation, trait association, and the prioritization of transcript-level events. By combining these capabilities, the framework moves beyond identifying isolated transcript changes toward understanding how transcript variation is organized across the transcriptome.

What Did We Discover in the Aging Human Brain?
To evaluate IsoGraph, we combined synthetic benchmarking with 16 aging analyses of adult human brain transcriptomic data: three brain regions from BrainSEQ and 13 from the Genotype-Tissue Expression (GTEx) project. Our findings highlight several important insights.
1. Coordinated Transcript Choice Reveals Information Beyond Gene Expression
IsoGraph identified age-associated patterns of transcript usage that persisted even after accounting for total gene abundance. This suggests that studying which RNA isoforms are expressed can reveal additional dimensions of brain aging that are not fully captured by measuring how much RNA a gene produces.
These associations also persisted after adjustment for estimated cellular composition in some analyses, although the results depended on the brain region examined.
2. Transcript-Usage Network Axes Transfer Across Cohorts
We found that IsoGraph modules were stable across subsets of donors. Importantly, network axes defined in one cohort retained their age associations when applied to another cohort, even though the exact genes assigned to individual modules differed.
These findings suggest that coordinated transcript-usage patterns are more reproducible as network axes than as fixed lists of genes, despite variability in the precise network structure recovered across datasets.
3. Network Context Helps Identify Transcript-Level Changes
IsoGraph can provide information beyond examining each gene independently. In three of six cohort-region analyses, module-level context predicted differential transcript usage in held-out donors beyond gene-wise evidence and conventional gene co-expression. The effect was modest and depended on the brain region, so network context complements gene-by-gene analysis without replacing it.
We also evaluated individual transcript events using several complementary sources of evidence, including transcript structural annotations, short-read splice-junction usage, independent long-read sequencing, and within-donor allelic analyses. Together, these analyses provided additional support for selected transcript-level signals identified through the framework.
4. Connecting Transcript Choice to Disease-Relevant Genetic Signals
Beyond aging, we investigated how IsoGraph findings relate to neurological and psychiatric disease-associated genetic loci. Through genetic colocalization analyses, we prioritized candidate transcript events, including an alternative terminal-exon event involving PRDM2 at a locus associated with amyotrophic lateral sclerosis (ALS) and a splice junction in SH3GL2 at a locus associated with Parkinson’s disease.
These findings highlight the potential of integrating coordinated transcript-usage networks with genetic evidence to generate more focused hypotheses about disease-relevant molecular mechanisms. The genetic support was specific to selected genes and transcript events and did not extend across entire modules. The prioritized events remain candidates for further investigation rather than established causal mechanisms.
5. Age-Associated Switches Often Differ in Untranslated Regions
We also asked what distinguishes the transcripts involved in age-associated switches. Across nine aging analyses, these transcript pairs were consistently enriched for differences in their untranslated regions (UTRs), about 1.25-fold relative to other transcript pairs from the same genes, while differences in coding sequence showed much weaker enrichment.
UTRs influence RNA stability, localization, and translation, so this pattern points to specific, testable hypotheses about what these switches may do. The annotations do not establish the functional effect of any individual event.
Benchmarking and Reproducibility
We also evaluated IsoGraph using synthetic datasets with known underlying signals. These benchmarks demonstrated that the value of transcript-usage and gene-abundance representations depends on the biological signal being studied.
IsoGraph preferentially recovered network structure when coordinated transcript usage was the underlying signal, while abundance-based representations provided advantages in abundance-dominated and RNA-degradation settings. This distinction is important: IsoGraph is designed to examine complementary forms of molecular variation, not to suggest that transcript usage should replace conventional gene expression analysis.
To support transparency and reproducibility, we have made the IsoGraph software and accompanying analysis resources available to the research community.
Why This Matters
Understanding the molecular processes associated with human brain aging requires examining more than whether genes are expressed at higher or lower levels. Our findings suggest that the coordinated selection of RNA isoforms is a measurable dimension of molecular variation that complements total gene expression.
IsoGraph provides a framework for studying this variation in the context of gene abundance, identifying coordinated modules, and following network-level signals back to individual transcript events that can be investigated further. Although our findings establish associations rather than causal relationships, they provide a foundation for investigating how transcript regulation contributes to biological aging and potentially to neurological disease.
Because IsoGraph runs on bulk RNA-seq, it also makes coordinated transcript choice accessible to network biology, turning existing cohorts into a resource for identifying and prioritizing disease-relevant transcript regulation.
We hope IsoGraph will be a useful resource for researchers studying alternative splicing, transcriptomics, systems biology, brain aging, and disease-associated molecular variation.
Explore the Preprint and Software
We’re excited to share IsoGraph with the broader scientific community and welcome feedback, discussion, and opportunities for collaboration.
- Read the preprint: IsoGraph resolves coordinated transcript choice and gene abundance in the aging human brain
- Explore IsoGraph on GitHub: github.com/heart-gen/IsoGraph
- Install the software: IsoGraph on PyPI
- Read the documentation: isograph.readthedocs.io
- Access the benchmarking resources: Zenodo benchmark archive
- Share the preprint: Download the bioRxiv QR code
- More software from the lab: HEART-GeN software page
Preprint authors: Alexis Bennett, Elisa Kain Johnson, and Kynon J. M. Benjamin.
This manuscript is currently available as a bioRxiv preprint and has not yet undergone peer review.
Advancing transcriptomics means understanding not only how much RNA genes produce, but also which transcripts they choose, and how those choices are coordinated across the genome.