Key findings

  • Threeevidentiary gaps currently prevent routine clinical implementation: uncertain incremental validity, incomplete mechanistic validity and largely unproven clinical utility
  • Incremental clinical utilityis the decisive standard, not model complexity or the number of molecular layers incorporated
  • Hypothesescomputationally inferred endotypes and mechanistic-AI treatment assignments should be regarded as testable hypotheses for prospective testing, not validated clinical subtypes or recommendations
  • ExploratoryOperon™ (GATC Health) is included as one illustrative example; its diabetes-specific predictive performance has not been independently established in peer-reviewed prospective studies

Abstract

Artificial intelligence (AI) creates an opportunity to move personalized diabetes medicine from classification of established risk factors toward prospective prediction of disease trajectories. This transition is particularly relevant to type 2 diabetes, in which heterogeneous metabolic, inflammatory, vascular, renal, hepatic, and myocardial mechanisms can produce substantially different risks despite similar conventional clinical phenotypes. Current markers including HbA1c, fasting glucose, blood pressure, estimated glomerular filtration rate, albuminuria, and standard lipid measures remain essential but incompletely capture early mechanistic risk, treatment-response heterogeneity, and progression toward cardiometabolic complications.

Multiomic technologies, i.e., genomics, epigenomics, transcriptomics, proteomics, metabolomics, lipidomics, microbiomics, and extracellular vesicle profiling, provide complementary views of inherited liability, regulatory state, pathway activation, metabolic flux, and inter-organ signaling. However, multiomic translation is limited by high dimensionality, temporal variability, tissue specificity, assay heterogeneity, and difficulty distinguishing causal drivers from downstream correlates. Mechanistic artificial intelligence (AI) offers a potential approach for integrating these data into biologically interpretable models capable of disease-trajectory prediction, endotype classification, and hypothesis-driven therapeutic stratification. Platforms should be distinguished from validated clinical decision systems; disease-specific applications require external and prospective validation before clinical utility can be established. This review proposes a framework for AI-enabled prediction of cardiometabolic disease in diabetes that integrates longitudinal clinical phenotyping, multiomics, mechanistic endotyping, and intervention mapping, while defining the validation, equity, and implementation requirements necessary for translation into precision medicine. This narrative review critically examines whether integration of longitudinal clinical phenotyping, multiomic profiling, and mechanistically informed computational modeling can improve cardiometabolic trajectory prediction beyond established clinical assessment and identifies the validation, generalizability, and implementation requirements necessary for clinical translation.

Introduction

Type 2 diabetes (T2D) is a heterogeneous metabolic disorder in which hyperglycemia arises within a broader network of cardiometabolic dysfunction1. Insulin resistance, β-cell dysfunction, adipose tissue inflammation, altered hepatic lipid metabolism, endothelial injury, kidney dysfunction, neurohormonal activation, and myocardial remodeling can interact over time to shape disease progression and complications2,3.. Consistent with this broader disease framework, the American Diabetes Association (ADA) Standards of Care emphasize systematic assessment of atherosclerotic cardiovascular disease (ASCVD), heart failure (HF), chronic kidney disease (CKD), and associated risk factors, with treatment selection increasingly informed by cardiovascular and renal benefit rather than glycemic control alone 4,5. Personalized medicine in T2D therefore extends beyond diagnosis and conventional risk factor classification toward anticipating how disease is likely to evolve in an individual patient6.

Clinically meaningful prediction requires distinguishing patients likely to progress from metabolic risk to overt diabetes or to specific cardiovascular, renal, hepatic, or microvascular complications, while also identifying the biological processes associated with those trajectories. The longer-term objective is to determine whether this biological information can refine prevention or treatment selection beyond established clinical phenotypes and guideline-based risk assessment. This requires moving from population-level associations toward longitudinal models that connect individual risk, underlying biology, and potentially modifiable disease trajectories.

This objective is consistent with the ADA/European Association for the Study of Diabetes (EASD) Precision Medicine in Diabetes Initiative, which identifies heterogeneity as a central challenge across diabetes prevention, diagnosis, treatment, prognosis, and monitoring7,8This narrative review critically examines whether integration of longitudinal clinical phenotyping, multiomic profiling, and mechanistically informed computational modeling can improve prediction of cardiometabolic disease trajectories in T2D beyond established clinical risk assessment. The central premise of this review is that the principal opportunity for artificial intelligence in precision diabetes medicine lies not in increasing predictive complexity alone, but in determining whether longitudinal clinical and molecular data can identify biologically interpretable risk states before irreversible cardiometabolic injury occurs. Particular emphasis is placed on whether multiomic profiling provides clinically meaningful biological resolution beyond conventional phenotyping and whether mechanistically informed computational models can translate these data into interpretable predictions of disease trajectory and therapeutic response. Throughout the review, predictive association is distinguished from mechanistic and causal inference, and the evidence required for translation is considered in terms of external validation, calibration, generalizability, prospective evaluation, and clinical utility.

Review Approach and Scope

This narrative review focuses on human evidence relevant to prediction of T2D progression and major cardiometabolic outcomes, with particular emphasis on cardiovascular, renal, hepatic, and heart-failure trajectories. Literature was identified through targeted searches of PubMed/MEDLINE and reference lists of relevant reviews, consensus statements, and primary studies, prioritizing contemporary human studies, longitudinal cohorts, externally validated prediction studies, and methodological guidance relevant to multiomics and artificial intelligence. Seminal earlier studies were included where necessary to establish conceptual or mechanistic context. Because the objective was critical synthesis rather than exhaustive evidence enumeration, studies were selected for their relevance to three prespecified themes: incremental information provided by multiomics, mechanistic interpretability of computational models, and requirements for clinical translation. This article is a narrative review and did not use a systematic-review protocol or quantitative meta-analysis. Individual omic modalities and cardiometabolic complications are therefore considered only insofar as they inform these two central themes rather than as comprehensive reviews of their respective fields.

Why cardiometabolic disease prediction is a precision-medicine problem

Conventional measures including HbA1c, blood pressure, lipid parameters, estimated glomerular filtration rate, and albuminuria remain indispensable for risk assessment but incompletely resolve the biological heterogeneity underlying cardiometabolic progression. Similar glycemic phenotypes can coexist with substantially different contributions from β-cell dysfunction, hepatic and adipose insulin resistance, dyslipidemia, inflammation, vascular injury, renal dysfunction, and myocardial stress7–11. Consequently, the relevant precision-medicine problem is not whether established clinical predictors should be replaced, but whether molecular information provides reproducible and actionable information beyond them.

This distinction establishes a stringent benchmark for AI-enabled prediction. Incremental discrimination alone is insufficient; a clinically useful model should improve calibration or risk classification, identify information relevant to a defined clinical decision, and demonstrate transportability and prospective utility. The value of molecular endotyping therefore depends on whether it changes prediction or management beyond appropriately specified clinical baseline models.

Multiomics as the biological substrate for disease prediction

Multiomics provides the biological resolution required to move from phenotype-based diabetes care to mechanism-based diabetes care. Genomics captures inherited liability for β-cell function, insulin secretion, adiposity, lipid metabolism, renal vulnerability, and vascular risk12,13. Epigenomics captures environmentally responsive regulatory states, including DNA methylation and chromatin remodeling linked to metabolic memory, inflammation, and long-term complication risk14. Transcriptomics reflects dynamic pathway activation, including immune, oxidative stress, endoplasmic-reticulum stress, mitochondrial, and endothelial programs15. Proteomics quantifies circulating mediators of inflammation, endothelial dysfunction, coagulation, complement activation, renal injury, and myocardial stress16. Metabolomics characterizes circulating or tissue metabolite abundance associated with amino-acid metabolism, lipid utilization, ketone metabolism, and mitochondrial substrate handling; these measurements should not be interpreted as direct measures of pathway flux without complementary dynamic approaches such as stable-isotope tracing17. Lipidomics resolves ceramides, sphingolipids, phospholipids, remnant-rich phenotypes, and lipotoxic signatures18. Microbiomics captures host–microbe interactions through bile acid metabolism, short-chain fatty acids, intestinal barrier signaling, and systemic inflammation19. These modalities interrogate complementary levels of biological organization: inherited susceptibility through genomics; regulatory and transcriptional state through epigenomic and transcriptomic profiling; pathway and signaling activity through proteomics; metabolic state through metabolomics and lipidomics; and host–microbial interactions through microbiome-derived features20–26. Their principal value for prediction is therefore not simply additive dimensionality, but the possibility of linking upstream susceptibility to intermediate molecular states and downstream clinical phenotypes. However, biological complementarity does not guarantee incremental predictive value, particularly when molecular layers are correlated or when established clinical variables already capture substantial prognostic information.

Evidence is emerging that multiomic prediction can improve diabetes risk models. Carrasco-Zanini and colleagues evaluated genome, proteome, metabolome, and clinical biomarkers for incident T2D prediction and showed that combined omics layers can add information beyond standard models, while also emphasizing that clinical utility depends on context and absolute risk27. More recently, Li and colleagues integrated epigenomics, proteomics, and metabolomics to identify diabetes-associated biomarkers and signaling pathways. The authors demonstrated the early-prediction model improved area under the receiver operating characteristic curve compared with a clinical baseline model and prioritized biomarkers with potential causal and therapeutic relevance28. These studies demonstrate that molecular profiling can add predictive information in selected settings, but they do not establish that broader multiomic integration will consistently outperform parsimonious clinical or single-omic models. Incremental performance should therefore be evaluated against prespecified clinical baselines, with attention to calibration, external validation, assay burden, and whether additional molecular information changes a clinically meaningful decision.Notwithstanding, multiomics creates analytical complexity. High-dimensional datasets are vulnerable to batch effects, missingness, site variability, assay drift, confounding, population imbalance, and overfitting. Blood-based omics may not fully represent liver, adipose tissue, skeletal muscle, pancreatic islets, myocardium, or vascular endothelium. Single timepoint sampling may confuse transient physiology with stable risk. Clinical translation therefore requires models that are longitudinal, externally validated, calibrated, interpretable, and linked to clear decisions.

Artificial Intelligence: From Statistical Risk Prediction to Mechanistic Inference

Artificial Intelligence (AI) and machine learning (ML) are increasingly applied to T2D diagnosis, prognosis, phenotyping, and treatment-response modeling. A recent review of ML in precision diabetes care and cardiovascular risk highlights the potential of AI to support diagnosis, prognostication, phenotyping, and treatment of diabetes and cardiovascular complications29. However, conventional ML can remain clinically limited if it optimizes discrimination without mechanistic interpretability30. For example, a model that predicts HF hospitalization but cannot explain whether risk is driven by renal congestion, myocardial metabolic dysfunction, obesity-related hemodynamics, inflammation, or ischemic substrate may be less useful for precision intervention30. AI-enabled personalized T2D medicine should therefore be developed and evaluated according to contemporary prediction-model standards, including transparent specification of data sources, predictors and outcomes, assessment of discrimination and calibration, evaluation of subgroup performance, and external validation31. Clinical evaluation should therefore proceed hierarchically, such that a model should demonstrate incremental predictive performance beyond established clinical models, adequate calibration and subgroup performance, interpretable linkage between predictors and biological hypotheses, relevance to a defined clinical action, and ultimately prospective evidence of clinical utility.Notably, these criteria represent distinct evidentiary levels. Predictive accuracy establishes neither biological mechanism nor causal actionability, and mechanistic interpretability does not demonstrate that perturbing an inferred pathway will alter clinical outcome. Mechanistic AI differs from purely predictive AI by embedding molecular and clinical data into pathway structures, causal hypotheses, and scenario simulations. Instead of asking only “Who is high risk?” mechanistic AI asks “Which pathways make this person high risk, and what intervention is predicted to modify that trajectory?” This distinction is central in diabetes, where multiple therapeutic classes act on partially overlapping but distinct pathways.

A Framework for AI-Enabled Prediction of Cardiometabolic Disease in Diabetes

An AI-enabled personalized diabetes prediction framework should begin with deep clinical phenotyping integrated with longitudinal outcome ascertainment. Core clinical variables may include age, sex, ancestry, socioeconomic context, family history, BMI, waist circumference, blood pressure, HbA1c, fasting glucose, continuous glucose metrics where available, lipids, apoB, eGFR, albuminuria, liver enzymes, MASLD/MASH status, medication history, diet, physical activity, sleep, smoking, and prior cardiovascular or renal disease. However, a secondary core set encompasses molecular characterization. Genomics can estimate inherited liability. Epigenomics can capture metabolic memory and exposure-related regulatory state. Proteomics can identify inflammation, renal injury, endothelial activation, and myocardial stress. Metabolomics and lipidomics can detect substrate overload, amino acid dysregulation, ceramide-related lipotoxicity, and mitochondrial stress. Microbiome and extracellular vesicle data can capture modifiable host–microbe interactions and inter-organ communication. Longitudinal modeling is layered across these cores, providing the contexts in which prediction can be time-to-event oriented rather than cross-sectional. Relevant endpoints include incident T2D in at-risk populations, progression from prediabetes to T2D, ASCVD events, HF hospitalization, CKD progression, albuminuria progression, MASLD/MASH progression, medication response, hypoglycemia, weight change, and treatment discontinuation. An additional layer is mechanistic endotyping, in which integrated molecular and clinical features are used to generate candidate pathway-enriched risk states, e.g., β-cell dysfunction, hepatic lipotoxicity, inflammatory vascular injury, renal vulnerability, or myocardial stress. These categories should be regarded as testable biological hypotheses rather than validated clinical subtypes. Their translational value would depend on demonstrating reproducibility across cohorts and, critically, treatment-effect heterogeneity: patients classified within a candidate endotype would need to show differential response to an intervention targeting the implicated biology. Thus, the framework connects prediction to an intervention hypothesis without assuming that computationally inferred endotypes are themselves clinically actionable (Figure 1).

Figure 1. Conceptual framework for AI-enabled prediction of cardiometabolic disease trajectories in diabetes. Prediction is treated as a translational hypothesis generator; clinical utility requires external validation, calibration, prospective testing, and monitoring for model drift.

Operon as an Illustrative Mechanistic AI Architecture

Operon™ (GATC Health, Irvine, California) is a proprietary computational platform designed to integrate heterogeneous molecular, clinical, and biological knowledge for systems-level hypothesis generation. It is included here as one illustrative example of the broader mechanistic-AI framework described above rather than as a validated clinical prediction system. Because its architecture is proprietary and diabetes-specific predictive performance has not been independently established in peer-reviewed prospective studies, platform-generated endotypes or intervention hypotheses should be considered exploratory. As with any comparable computational framework, clinical translation would require independent external validation, calibration, prospective testing, and demonstration of incremental clinical utility beyond established approaches.

From Prediction to Action: Personalized Prevention and Treatment

The clinical value of AI-enabled prediction ultimately depends on actionability. A model that improves discrimination without altering prevention, monitoring, or treatment decisions may have limited incremental value over conventional risk assessment.

Prediction-guided treatment

Current diabetes treatment already incorporates precision principles through selection of therapies according to established cardiovascular, renal, metabolic, and weight-related phenotypes32. The unresolved question is whether molecular endotyping can improve treatment selection beyond these clinically observable characteristics. For example, molecular signatures could theoretically identify treatment-response heterogeneity within patients who otherwise share similar clinical indications, but such applications require demonstration of treatment-by-endotype interactions rather than associations between baseline biomarkers and outcomes alone. The appropriate evidentiary standard is therefore comparative: an endotype-guided strategy should improve treatment selection, safety, or clinical outcomes relative to guideline-based care using routinely available information. Until such evidence exists, mechanistic-AI treatment assignments should be regarded as hypotheses for prospective testing rather than recommendations for clinical management.

Prediction-Guided Prevention. The same evidentiary principle applies to prevention. Molecular and computational profiling may ultimately distinguish individuals with similar conventional risk but different biological trajectories, potentially enabling more targeted preventive strategies. However, improved biological resolution has clinical value only if it identifies risk earlier or more accurately than established approaches and leads to an intervention that improves outcomes. Prospective intervention studies, rather than retrospective classification performance, are therefore required to establish precision-prevention utility.

AI-Enabled Clinical Trials and Real-World Translation

Before AI-enabled disease prediction can become a component of routine diabetes care, its value should be established prospectively through trials designed to test not only predictive accuracy but whether prediction changes clinically meaningful decisions and outcomes. A major limitation of T2D trials is heterogeneous enrollment. Patients with the same diagnostic label may have different biology, producing diluted effect sizes and unclear responder subgroups. AI-enabled endotyping can support enriched trial design by selecting patients whose biological pathways match the intervention mechanism. For example, a lipotoxicity-targeting therapy should be tested in patients with lipidomic and hepatic flux evidence of lipotoxic risk. An anti-inflammatory strategy should be tested in patients with inflammatory proteomic and endothelial activation signatures. A β-cell protective therapy should be tested in patients with preserved but stressed β-cell function. Real-world deployment requires governance. Emerging evidence and evolving regulatory frameworks indicate that clinically deployable AI models require transparent reporting of uncertainty, missingness, calibration, and subgroup performance to support interpretability and reproducibility across populations and care settings31. Longitudinal monitoring and recalibration are also likely necessary because model performance may drift over time as therapeutic standards, population characteristics, and clinical practices evolve. Current consensus across translational AI research further supports implementation of these systems as clinician-supervised decision-support frameworks rather than autonomous “black-box” determinants of care, particularly in complex diseases such as T2D where mechanistic heterogeneity and multimorbidity substantially influence treatment decisions and outcomes. Crucially, enrichment based on a computational endotype should be prospectively prespecified and evaluated for an endotype-by-treatment interaction; post hoc identification of responders would provide substantially weaker evidence of predictive utility.

Equity, Generalizability, and Responsible Deployment

Personalized medicine can worsen disparities if developed on narrow datasets or deployed only in high-resource settings. Multiomic and AI models often underrepresent non-European ancestries, rural populations, lower-income groups, and patients with multimorbidity. Polygenic risk scores may perform less accurately in populations underrepresented in discovery cohorts. Multiomic assays may be expensive, and serial molecular profiling may be inaccessible. Equitable personalized diabetes medicine requires diverse training and validation cohorts, transparent subgroup performance reporting, cost-effectiveness evaluation, and tiered implementation models. A pragmatic pathway might begin with clinical and laboratory AI models, followed by targeted molecular profiling for patients whose risk remains uncertain or whose treatment decisions would change33. The objective should therefore be the minimum set of clinical and molecular data that produces a reproducible and clinically meaningful improvement in decision-making.

Limitations and Research Priorities

Three evidentiary gaps currently prevent routine clinical implementation. First, incremental validity remains uncertain: many AI and multiomic models have not demonstrated robust improvement over appropriately specified clinical comparators in independent populations. Second, mechanistic validity remains incomplete because molecular associations and biologically interpretable model outputs do not establish causal pathways or treatment responsiveness. Third, clinical utility remains largely unproven because few studies demonstrate that model-guided decisions improve patient outcomes, efficiency, or equity. These limitations are compounded by assay variability, population underrepresentation, cost, workflow requirements, and model drift.

Future research should therefore prioritize prospective longitudinal validation, direct comparison of clinical-only versus clinical-plus-molecular models, prespecified testing of endotype-by-treatment interactions, and evaluation of calibration, transportability, cost-effectiveness, and subgroup performance. This sequence would move the field from increasingly complex prediction toward evidence that mechanism-informed prediction changes care.

Conclusions

The next frontier in personalized diabetes medicine is not simply more accurate classification of established disease, but earlier and more biologically informative prediction of disease trajectories. Type 2 diabetes provides a particularly important use case because similar clinical phenotypes can arise from different combinations of β-cell dysfunction, insulin resistance, hepatic lipotoxicity, inflammation, vascular injury, renal vulnerability, and myocardial stress. Multiomic profiling offers increasingly detailed measurement of these processes, but its clinical value will depend on methods capable of integrating high-dimensional signals into interpretable and actionable models. Mechanistic AI provides a potential framework for this transition by linking multidimensional biomarkers to biological pathways, pathways to future risk, and predicted risk to prevention or therapeutic hypotheses. The decisive standard is therefore not model complexity or the number of molecular layers incorporated, but incremental clinical utility. Mechanism-informed prediction will advance precision diabetes medicine only if it provides reproducible information beyond established clinical models, remains calibrated and transportable across populations, identifies biologically plausible and testable intervention hypotheses, and ultimately improves patient outcomes. The field should therefore progress from retrospective prediction toward prospective tests of whether biologically informed risk stratification can alter disease trajectories.

Declarations

Conflicts of interest
All authors are employees of and/or consultants to GATC Health Corp. and may have professional or financial relationships with the company. GATC Health Corp. develops the Operon™ platform discussed in this manuscript. These relationships are disclosed for transparency. The manuscript is a narrative review and perspective based on the cited scientific literature; descriptions of Operon™ are presented as an illustrative translational framework and are not intended to represent independently validated evidence of clinical efficacy or utility.
Funding
No external funding was received for the preparation of this manuscript.
Author contributions
Conceptualization: Jonathan R.T. Lakey and Krista Casazza; Methodology and scientific framework: Krista Casazza and Jonathan R.T. Lakey; Literature review and interpretation: Krista Casazza and Jonathan R.T. Lakey; Writing-original draft preparation: Krista Casazza and Jonathan R.T. Lakey; Writing-review and editing: Jonathan R.T. Lakey, Krista Casazza, Ian Jenkins, Waldemar Lernhardt, Valentina Savich, and Jayson Uffens; Operon™ platform expertise and technical input: Ian Jenkins, Waldemar Lernhardt, Valentina Savich, and Jayson Uffens; Supervision: Jonathan R.T. Lakey. All authors reviewed and approved the final manuscript.

References

  1. Lu X, Xie Q, Pan X, et al. Type 2 diabetes mellitus in adults: pathogenesis, prevention and therapy. Signal Transduct Target Ther. 2024;9(1):262. doi:10.1038/s41392-024-01951-9
  2. Ortiz-Martínez M, González-González M, Martagón AJ, Hlavinka V, Willson RC, Rito-Palomares M. Recent Developments in Biomarkers for Diagnosis and Screening of Type 2 Diabetes Mellitus. Curr Diab Rep. 2022;22(3):95-115. doi:10.1007/s11892-022-01453-4
  3. Ahlqvist E, Prasad RB, Groop L. Subtypes of Type 2 Diabetes Determined From Clinical Parameters. Diabetes. 2020;69(10):2086-2093. doi:10.2337/dbi20-0001
  4. American Diabetes Association Professional Practice Committee for Diabetes*. 4. Comprehensive Medical Evaluation and Assessment of Comorbidities: Standards of Care in Diabetes-2026. Diabetes Care. 2026;49(Supplement_1):S61-S88. doi:10.2337/dc26-S004
  5. American Diabetes Association Professional Practice Committee for Diabetes*. 10. Cardiovascular Disease and Risk Management: Standards of Care in Diabetes-2026. Diabetes Care. 2026;49(Suppl 1):S216-S245. doi:10.2337/dc26-S010
  6. Franks PW, Sargent JL. Diabetes and obesity: leveraging heterogeneity for precision medicine. Eur Heart J. 2024;45(48):5146-5155. doi:10.1093/eurheartj/ehae746
  7. Chung WK, Erion K, Florez JC, et al. Precision Medicine in Diabetes: A Consensus Report From the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD). Diabetes Care. 2020;43(7):1617-1635. doi:10.2337/dci20-0022
  8. Nolan JJ, Kahkoska AR, Semnani-Azad Z, et al. ADA/EASD Precision Medicine in Diabetes Initiative: An International Perspective and Future Vision for Precision Medicine in Diabetes. Diabetes Care. 2022;45(2):261-266. doi:10.2337/dc21-2216
  9. Quispe R, Martin SS, Michos ED, et al. Remnant cholesterol predicts cardiovascular disease beyond LDL and ApoB: a primary prevention study. Eur Heart J. 2021;42(42):4324-4332. doi:10.1093/eurheartj/ehab432
  10. Baidya D, Hanumanpratap Singh Kshatri A, Zerzan EK, Menon GJ, Sawale M, Subhash Bhalodiya S. The Metabolic Heart: Reframing Heart Failure With Preserved Ejection Fraction as a Systemic Cardio-Metabolic Syndrome. Cureus. 2026;18(1):e100590. doi:10.7759/cureus.100590
  11. Abramov D, Mathew RO, Fordan SV. Heart Failure and Comorbidities (Chronic Kidney Disease, Diabetes, Obesity) Management: A Multidisciplinary Approach. Cardiorenal Med. 2026;16(1):95-112. doi:10.1159/000550503
  12. Chami N, Wang Z, Svenstrup V, et al. Genetic subtyping of obesity reveals biological insights into the uncoupling of adiposity from its cardiometabolic comorbidities. Nat Med. 2025;31(11):3801-3812. doi:10.1038/s41591-025-03931-0
  13. Billings LK, Jablonski KA, Pan Q, et al. Increased Genetic Risk for β-Cell Failure Is Associated With β-Cell Function Decline in People With Prediabetes. Diabetes. 2024;73(8):1352-1360. doi:10.2337/db23-0761
  14. Wu YL, Lin ZJ, Li CC, et al. Epigenetic regulation in metabolic diseases: mechanisms and advances in clinical study. Signal Transduct Target Ther. 2023;8(1):98. doi:10.1038/s41392-023-01333-7
  15. Rivera-Alvarez I, Vázquez-Lizárraga R, Mendoza-Viveros L, et al. Transcriptional dynamics in type 2 diabetes progression is linked with circadian, thermogenic, and cellular stress in human adipose tissue. Commun Biol. 2025;8(1):398. doi:10.1038/s42003-025-07709-5
  16. Salman O, Zhao L, Cohen JB, et al. Proteomic Correlates and Prognostic Significance of Kidney Injury in Heart Failure With Preserved Ejection Fraction. J Am Heart Assoc. 2024;13(17):e033660. doi:10.1161/JAHA.123.033660
  17. Arneth B, Arneth R, Shams M. Metabolomics of Type 1 and Type 2 Diabetes. Int J Mol Sci. 2019;20(10):2467. doi:10.3390/ijms20102467
  18. Ağagündüz D, Icer MA, Yesildemir O, Koçak T, Kocyigit E, Capasso R. The roles of dietary lipids and lipidomics in gut-brain axis in type 2 diabetes mellitus. J Transl Med. 2023;21(1):240. doi:10.1186/s12967-023-04088-5
  19. Zhang Y, Wang DD. Gut microbiome in type 2 diabetes: insights from metagenomics, multi-omics, and diet-microbe interactions. Gut Microbes. 2026;18(1):2644682. doi:10.1080/19490976.2026.2644682
  20. Chami N, Wang Z, Svenstrup V, et al. Genetic subtyping of obesity reveals biological insights into the uncoupling of adiposity from its cardiometabolic comorbidities. Nat Med. 2025;31(11):3801-3812. doi:10.1038/s41591-025-03931-0
  21. Billings LK, Jablonski KA, Pan Q, et al. Increased Genetic Risk for β-Cell Failure Is Associated With β-Cell Function Decline in People With Prediabetes. Diabetes. 2024;73(8):1352-1360. doi:10.2337/db23-0761
  22. Wu YL, Lin ZJ, Li CC, et al. Epigenetic regulation in metabolic diseases: mechanisms and advances in clinical study. Signal Transduct Target Ther. 2023;8(1):98. doi:10.1038/s41392-023-01333-7
  23. Rivera-Alvarez I, Vázquez-Lizárraga R, Mendoza-Viveros L, et al. Transcriptional dynamics in type 2 diabetes progression is linked with circadian, thermogenic, and cellular stress in human adipose tissue. Commun Biol. 2025;8(1):398. doi:10.1038/s42003-025-07709-5
  24. Salman O, Zhao L, Cohen JB, et al. Proteomic Correlates and Prognostic Significance of Kidney Injury in Heart Failure With Preserved Ejection Fraction. J Am Heart Assoc. 2024;13(17):e033660. doi:10.1161/JAHA.123.033660
  25. Ağagündüz D, Icer MA, Yesildemir O, Koçak T, Kocyigit E, Capasso R. The roles of dietary lipids and lipidomics in gut-brain axis in type 2 diabetes mellitus. J Transl Med. 2023;21(1):240. doi:10.1186/s12967-023-04088-5
  26. Zhang Y, Wang DD. Gut microbiome in type 2 diabetes: insights from metagenomics, multi-omics, and diet-microbe interactions. Gut Microbes. 2026;18(1):2644682. doi:10.1080/19490976.2026.2644682
  27. Carrasco-Zanini J, Pietzner M, Wheeler E, Kerrison ND, Langenberg C, Wareham NJ. Multi-omic prediction of incident type 2 diabetes. Diabetologia. 2024;67(1):102-112. doi:10.1007/s00125-023-06027-x
  28. Li W, Cheng Y, Cui A, et al. Multiomics Integration of Epigenetics, Proteomics, and Metabolomics Identifies Putative Drug Targets and Improves Early Prediction for Diabetes. Diabetes. 2025;74(12):2418-2431. doi:10.2337/db25-0354
  29. Oikonomou EK, Khera R. Machine learning in precision diabetes care and cardiovascular risk prediction. Cardiovasc Diabetol. 2023;22(1):259. doi:10.1186/s12933-023-01985-3
  30. Mateussi N, Rogers MP, Grimsley EA, et al. Clinical Applications of Machine Learning. Ann Surg Open Perspect Surg Hist Educ Clin Approaches. 2024;5(2):e423. doi:10.1097/AS9.0000000000000423
  31. Collins GS, Moons KGM, Dhiman P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. doi:10.1136/bmj-2023-078378
  32. American Diabetes Association Professional Practice Committee for Diabetes*. 9. Pharmacologic Approaches to Glycemic Treatment: Standards of Care in Diabetes-2026. Diabetes Care. 2026;49(Suppl 1):S183-S215. doi:10.2337/dc26-S009
  33. Bai B, Liu X, Li H. Federated multimodal AI for precision-equitable diabetes care. Front Digit Health. 2025;7:1678047. doi:10.3389/fdgth.2025.1678047

Article information

Authors

How to cite

Lakey, J. R. T., Jenkins, I., et al., 2026. AI-Enabled Prediction of Cardiometabolic Disease in Diabetes: Integrating Multiomics, Mechanistic Modeling, and Precision Medicine. Medical Research Archives, [online] 14(10). https://doi.org/10.18103/mra.2026.0639

Record

Journal
Medical Research Archives
Section
Review Articles
Published
31 October 2026
ISSN
2375-1924
Downloads
58

Copyright and licence

Published by the European Society of Medicine under a Creative Commons Attribution 4.0 International licence (CC BY 4.0), which permits unrestricted use, distribution and reproduction in any medium, provided the original authors and source are credited.