ISMB 2026 – iRNA COSI Meeting

Alternative RNA splicing exponentially expands the functional proteome, yet its systematic dysregulation across cancer and aging remains incompletely understood. Our work integrates multi-modal genomic approaches to decode how splicing programs are remodeled during aging, malignant transformation, and tumor immune evasion. Using long-read RNA-sequencing of human breast and lung tumors, we have mapped the landscape of full-length cancer isoforms, uncovering thousands of novel transcripts that are often missed by short-read approaches. Building on this isoform atlas, we reveal that aging drives broad splicing remodeling in mammary epithelial cells, and that select tumor-associated isoform signatures are in fact age-driven, positioning splicing alterations as early preneoplastic events. We further define the regulatory architecture controlling these splicing programs, demonstrating that post-transcriptional autoregulatory mechanisms within splicing factors are selectively disrupted in tumors, representing novel intervention points. Finally, we show that specific splicing events have direct clinical relevance: intron retention in immune-regulatory genes predicts patient response to checkpoint inhibitor immunotherapy, and splice-switching oligonucleotides that correct tumor-associated isoforms exert selective anti-proliferative effects across cancer models. Together, this work defines splicing dysregulation as a molecular hallmark of aging and cancer, and identifies novel isoforms and splicing regulators as actionable biomarkers and therapeutic targets.

The overarching vision of our research program is to construct predictive models that explain how cells determine their protein abundance. Achieving this goal involves two major components: (1) higher-resolution and higher-precision measurements of gene expression modalities, and (2) computational and theoretical advancements capable of integrating these quantitative measurements into cohesive, predictive frameworks. Towards these goals, we have compiled measurements of translation from more than 3,500 experiments and introduced the concept of translation efficiency covariation (TEC), revealing that transcripts associated with shared biological functions and those that are members of the same protein complexes exhibit TEC. We then leveraged our expansive compendium of translation efficiency measurements to develop a deep neural network model, called RiboNN, capable of predicting mRNA translation rates across numerous cell types based solely on the full-length mRNA sequence. RiboNN can evaluate the impact of genetic variants in the human population, providing insight into diseases driven by abnormal mRNA translation. This approach has implications for bioengineering applications, genetic diagnostics as well as the design and optimization of mRNA therapies.

In vitro evolution has greatly expanded the functional repertoire of RNA and highlights opportunities to explore beyond natural natural nucleic acid chemistry. Here, I will present our efforts to extend RNA biology using non-natural nucleic acids (XNAs). We focused on an XNA-SELEX platform that enables in vitro evolution of XNAs with programmable binding and catalytic activities. This approach has yielded XNA molecules capable of precise RNA targeting and protein modulation in complex biological contexts. Our work further demonstrates how alternative backbone chemistries can access functional properties not readily available to canonical RNA. Together, these results establish XNAs as a versatile molecular toolkit for probing and modulating RNA and protein function, and provide a general framework for extending nucleic acid structures and functions beyond natural systems.

Despite their pivotal roles in gene regulation, the precise timing of mRNA modifications during biogenesis has remained elusive due to the lack of methods that can simultaneously resolve modification status and RNA processing states at the single-molecule level. Here we describe two complementary nanopore-based approaches to establish a temporal framework for mRNA modification. First, we developed SWARM (Single-molecule Workflow for Analysing RNA Modifications) (Prodic et al. 2026), an AI-based tool that overcomes the high false-positive rates plaguing modification detection. By employing a training strategy that accounts for signal crosstalk from non-target modifications and incorporates orthogonally validated cellular signals, SWARM enables high-precision detection of m6A and pseudouridine (Ψ) at single-nucleotide and single-molecule resolution and outperforms existing tools on rigorous cellular benchmarks. Second, to place these modifications in time, we created dFORCE (direct Fractionated Observation of RNA Coupled to Elongation) (Sethi et al. 2025), a technique that separates transcriptionally engaged pre-mRNA from mature, polyadenylated transcripts, allowing us to pinpoint the order of processing events on a transcriptome-wide scale. Applying these technologies across cell lines, mammalian tissues, and targeted perturbation experiments with splicing and modification enzyme inhibitors reveals two key temporal principles of RNA modification. First, we find that m6A deposition substantially precedes splicing catalysis, with m6A accumulating near-cytoplasmic levels on unspliced pre-mRNA. Crucially, the canonical suppression of m6A near exon boundaries is already in place before splicing and the deposition of the exon junction complex, challenging prevailing models of how m6A topology is established. Second, we discover a splicing-shaped mode of Ψ deposition in which TRUB1-mediated pseudouridylation near exon boundaries preferentially occurs after exon-exon ligation, driven by local RNA structure stabilisation in spliced transcripts. We validate this mechanism using both nascent RNA sequencing and in vitro biochemical assays. Together, these findings suggest that m6A and Ψ modifications follow distinct temporal trajectories during mRNA maturation, arguing against a globally coordinated epitranscriptomic code and instead pointing to a more intricate, stepwise regulatory logic.  and challenge models of widespread coordination between these marks. By integrating high-precision detection with temporal resolution, our integrated approach provides a robust framework for decoding the dynamic life of RNA.