Diseases rarely unfold in a smooth, predictable line. For patients and clinicians alike, the most consequential moments in an illness are often the abrupt ones: the day a stable tumor suddenly turns aggressive, the hour a mild respiratory infection tips into life-threatening inflammation, the point at which a chronic condition crosses into acute crisis. Scientists have long suspected that these dramatic shifts, known in complex-systems theory as critical transitions, are foreshadowed by measurable changes in the web of interactions among genes. A gene regulatory network under stress does not collapse without warning; it sends out signals, subtle rearrangements in how genes influence one another, that accumulate as the system approaches a tipping point. The challenge has been that these signals are fleeting, noisy, and scattered across the different ways biologists measure gene activity. Now a team of researchers at City University of Hong Kong and the Hangzhou Institute for Advanced Study has unveiled a computational framework designed to catch those warnings before the cliff edge arrives.
The new tool, called CRISGI, is described in the open-access journal Genome Biology by Chengshang Lyu, Anna Jiang, Ka Ho Ng, Xiaoyu Liu, Xiaoping Liu, and corresponding author Lingxi Chen. Its central innovation is a shift in focus: instead of asking which individual genes or broad modules of genes are behaving unusually during disease progression, CRISGI models the dynamics of gene-gene interactions themselves, tracking how the relationships between pairs of genes destabilize as a biological system approaches a critical transition. This interaction-level view matters because complex diseases are rarely the product of a single rogue gene. They emerge from networks of regulatory relationships, and the earliest signs of an impending state change may live in the weakening or strengthening of specific connections rather than in the expression levels of any one gene.
Technically, CRISGI tackles three shortcomings the authors identified in earlier critical-transition methods. First, most existing tools rank candidate signals in an unsupervised fashion, scanning genes or modules for statistical patterns of criticality without anchoring those patterns to specific biological phenotypes or specific observations. CRISGI instead computes what the authors call critical transition scores at the level of individual interactions and delivers both phenotype-level and observation-level rankings through rank enrichment, meaning it can highlight the gene interactions whose transition dynamics are most characteristic of a particular disease state or a particular sample. Second, earlier approaches often relied on unweighted enrichment strategies, treating all hits as equal; CRISGI incorporates the magnitude and consistency of the transition signals across interactions, sharpening the separation between true early-warning markers and statistical noise. Third, and perhaps most importantly for the modern genomics landscape, the framework is deliberately multimodal, engineered to operate on bulk transcriptomics, which measures average expression across millions of cells in a tissue; single-cell transcriptomics, which resolves expression cell by cell; and spatial transcriptomics, which preserves the physical location of gene expression within tissue architecture.
The multimodal design is not a cosmetic feature. Bulk data is abundant and clinically mature, but it blurs the heterogeneity that drives disease. Single-cell data exposes that heterogeneity but can be expensive and technically demanding. Spatial transcriptomics adds the crucial dimension of location, letting researchers see whether a suspicious interaction is active across an entire tissue or concentrated in a specific microenvironment. By modeling interaction-level critical transition dynamics across all three data types with a shared conceptual framework, CRISGI allows researchers to triangulate: a critical interaction detected in bulk cohort data can be traced to particular cell populations in single-cell data and then localized to specific anatomical regions in spatial maps. The authors argue that this cross-modal consistency is precisely what earlier tools, built for one data type at a time, have failed to provide.
To establish that the method works, the team first subjected CRISGI to in silico benchmarks, simulated datasets in which the location and timing of critical transitions are known by construction. On these benchmarks, CRISGI outperformed existing critical-transition methods, correctly identifying which interactions were approaching instability and when the system was poised to flip. The benchmarks matter because critical-transition detection is inherently difficult to validate: in real diseases, the tipping point is often identified only in retrospect, long after the early-warning window has closed. Simulation offers a controlled proving ground where ground truth is defined, and CRISGI’s performance there gave the authors confidence to move to real biological data.
The first major real-world application came in H3N2 influenza, a virus whose progression from mild symptoms to severe illness depends on precisely the kind of abrupt host-response shifts that critical-transition theory describes. Analyzing transcriptomic time-course data from infected patients, CRISGI prioritized 128 gene interactions that predicted the onset of symptoms, and the team then tested these predictions against eight external validation datasets, an unusually thorough robustness check. The interactions identified before symptom onset were not random: they formed a coherent, reproducible signature of the host immune system’s regulatory machinery being reorganized in preparation for the transition from asymptomatic infection to active disease. In practical terms, this suggests that the molecular groundwork for symptomatic influenza is laid in advance and is detectable, if one knows where to look, at the level of gene-gene interactions rather than individual gene expression.
Cancer offered a second, and arguably more clinically consequential, testing ground. Mining data from The Cancer Genome Atlas, the large international resource that pairs tumor transcriptomics with patient survival information, the researchers uncovered stage-specific gene interactions linked to patient survival across multiple TCGA cohorts. The finding underscores a subtle but important point: the gene interactions that matter may differ from one stage of disease to another, so a single static biomarker list may miss the interactions that are actually driving progression at a given moment. By modeling transitions dynamically rather than statically, CRISGI could surface the interactions whose destabilization coincided with each stage and with survival outcomes, providing a more faithful picture of how tumor biology shifts over the course of the disease.
Two detailed case studies illustrate the kind of mechanistic hypotheses the tool can generate. In colorectal cancer, single-cell analysis highlighted interactions between CDK genes, the cyclin-dependent kinases that drive cell-cycle progression, and FOXO transcription factors, which regulate cell survival and proliferation and are themselves modulated by cell-cycle dynamics. The identification of CDK–FOXO interactions as critical transition signals in colorectal cancer cells suggests a testable proposition: that the crosstalk between cell-cycle machinery and stress-response transcription factors reaches an unstable state as tumor cells progress, offering a potential point of therapeutic intervention. In breast cancer, spatial transcriptomic analysis revealed LUM-centric interactions, networks of relationships organized around the gene lumican, concentrated in invasive regions of breast tumors. Lumican encodes an extracellular matrix protein, and its centrality in the critical-transition landscape of invasive tumor territories points to a role for matrix remodeling in the abrupt shift from contained to invasive growth. Both findings are framed by the authors as hypotheses for experimental follow-up, not conclusions, but they demonstrate how a purely computational analysis can point experimentalists toward specific molecular players and specific tissue locations.
The broader significance of the work lies in what it offers to the field of network medicine. Critical-transition theory, borrowed from physics and ecology, has been gaining traction in biology because so many disease processes, from epileptic seizures to cytokine storms to tumor progression, exhibit hallmark signs of tipping-point behavior. Tools that can read those signs directly from gene expression data, across the full spectrum of modern transcriptomic platforms, could change how diseases are monitored: instead of diagnosing a state change after it has happened, clinicians could, in principle, receive an early warning while intervention is still possible. The Hong Kong team emphasizes that CRISGI is delivered as software, making it available to other groups, and its open-access publication means the benchmarks and workflows can be scrutinized and extended by the community. The work was supported by the National Natural Science Foundation of China, the Research Grants Council of Hong Kong, the CityU-Tung Biomedical Sciences Centre, and a CityUHK Start-Up Grant, with open access partially funded by the City University of Hong Kong’s Open Access Publishing Fund.
Limitations and future directions remain, as with any methodological advance. The framework’s predictions are statistical inferences from observational data, and the 128 symptom-predictive interactions in influenza and the CDK–FOXO and LUM-centric findings in cancer will need experimental validation in cell and animal models before they inform clinical practice. But the study establishes a template that many labs are likely to follow: treat disease progression as a dynamical system, monitor the stability of gene-gene interactions rather than individual genes, and integrate bulk, single-cell, and spatial measurements to localize where and when the system is about to tip. If the approach proves robust across additional diseases and datasets, the transient, interaction-level signals that have eluded researchers may become a routine part of the molecular early-warning toolkit, transforming the sudden events that define so much of medicine from surprises into anticipated, and potentially preventable, transitions.
Subject of Research: Detection of interaction-level critical transitions in gene regulatory networks across bulk, single-cell, and spatial transcriptomics to predict disease progression.
Article Title: Charting critical transient gene interactions in disease progression across bulk, single-cell, and spatial transcriptomics
Article References: Lyu, C., Jiang, A., Ng, K. H., Liu, X., Liu, X., & Chen, L. (2026). Charting critical transient gene interactions in disease progression across bulk, single-cell, and spatial transcriptomics. Genome Biology. https://doi.org/10.1186/s13059-026-04265-x
Image Credits: AI Generated
DOI: 10.1186/s13059-026-04265-x
Keywords: critical transitions, gene regulatory networks, CRISGI, transcriptomics, single-cell analysis, spatial transcriptomics, H3N2 influenza, TCGA cohorts, colorectal cancer, breast cancer, CDK-FOXO interactions, disease progression
Cite Scienmag News
APA MLA Chicago
Juliet Wilcox. (September 21, 2026). New Computational Tool Flags the Tipping Points Where Genes Turn Disease Dangerous. Scienmag. https://scienmag.com/new-computational-tool-flags-the-tipping-points-where-genes-turn-disease-dangerous/
Juliet Wilcox. “New Computational Tool Flags the Tipping Points Where Genes Turn Disease Dangerous.” Scienmag, 21 September 2026, https://scienmag.com/new-computational-tool-flags-the-tipping-points-where-genes-turn-disease-dangerous/. Accessed 21 September 2026.
Juliet Wilcox. “New Computational Tool Flags the Tipping Points Where Genes Turn Disease Dangerous.” Scienmag. September 21, 2026. https://scienmag.com/new-computational-tool-flags-the-tipping-points-where-genes-turn-disease-dangerous/
Copy citation Download RIS
Tags: breast cancerCDK-FOXO interactionsColorectal cancerCRISGIcritical transitionsdisease progressiongene regulatory networksH3N2 influenzasingle-cell analysisSpatial transcriptomicsTCGA cohortsTranscriptomics
