Abstract
The human gut microbiome represents a complex and dynamic therapeutic target whose effective interrogation requires system-level analytical approaches beyond single-omics or reductive methods. This mini-review synthesizes recent advances in computational modeling and multi-omics integration relevant to the development of predictive, patient-tailored microbiome therapies. We critically assess the analytical strengths and limitations of genome-scale metabolic models (GEMs); generalized Lotka–Volterra and ODE-based community models; agent-based simulations; and statistical machine-learning frameworks and examine how their integration with metagenomics, metatranscriptomics, metaproteomics, and metabolomics can help bridge microbial functional potential with clinically relevant phenotypes. Representative applications–including MintTea for disease module identification, gNOMO2 for integrative microbiome profiling, and AGORA-based community metabolic modeling–illustrate the translational scope of these frameworks across inflammatory, metabolic, and infectious disease contexts. Hybrid ML–GEM frameworks have not yet been directly applied to FMT outcome prediction; however, the mechanistic principles underlying both approaches – metabolic compatibility modeling and data-driven responder stratification – suggest a compelling direction for future investigation, contingent on prospective validation in adequately powered and independent clinical cohorts. Persistent methodological challenges–such as data heterogeneity, batch effects across sequencing platforms, incomplete multi-omics coverage, and limited interpretability of complex machine-learning models–are being actively addressed through standardized preprocessing pipelines, explainable Artificial intelligence (AI) strategies, and federated analytics. While federated approaches enable privacy-preserving, multi-institutional model training, they introduce additional constraints related to non-identically distributed data, communication overhead, and uneven computational capacity. Overall, the convergence of mechanistic modeling, data-driven learning, and distributed analytical infrastructures may assist in advancing microbiome research from a largely correlational perspective toward mechanistic and ultimately prescriptive frameworks for precision microbiome medicine.
1 Introduction
The human microbiome is recognized as a key regulator of host physiology, influencing metabolism, immunity, and neuroendocrine signaling across multiple body sites (). The microbiome is a highly complex, dynamic, and context-dependent ecosystem with tightly connected community composition, functional potential, and host responses (). Microbiome dysbiosis is microbial community imbalance that has been associated with diverse conditions, including inflammatory bowel disease (IBD), colorectal cancer (CRC), cardiometabolic disorders, and neurological diseases, highlighting the microbiome as both a biomarker source and a potential therapeutic target (). Systems biology provides a robust conceptual and methodological framework to address the microbiome complexity by integrating diverse multi-omics data, such as genomics, transcriptomics, metabolomics, and proteomics, with computational modeling, facilitating a comprehensive characterization of microbial communities and their interactions with the host (). This approach has the potential to transform microbiome research from correlational relationships to mechanistic understanding and practical interventions, in accordance with precision medicine’s focus on tailored therapies (). Recent advances, including genome-scale metabolic models (GEMs), dynamic community simulations, and machine learning-driven data fusion, have enabled predictive modeling of microbiome responses to diet, probiotics, or pharmaceuticals (; ).
This mini-review is a focused synthesis of certain computational and multi-omic methods that are most relevant to gut microbiome research. It rigorously assesses advanced computational modeling frameworks and multi-omics integration methodologies within microbiome systems biology. We specifically analyze the advantages and intrinsic constraints of constraint-based metabolic models (FBA/GEMs), population-level dynamic models (ODEs), and individual-based simulations (ABMs) within the framework of gut microbial ecology. By integrating these mechanistic frameworks with multi-omics data, we establish a pathway for the advancement of microbiome-based therapeutics tailored to specific host profiles, thereby potentially facilitating clinical translation.
Although existing methodological resources and toolkits have established foundational infrastructure for constraint-based modeling, and prior reviews have examined multi-omics integration across diverse disease contexts (; ; ), our review addresses a distinct gap in three specific respects: (1) it systematically characterizes the integration of emerging ML–GEM hybrid frameworks into clinical translation workflows, emphasizing both their mechanistic utility and current validation constraints; (2) it identifies federated analytics as a principled solution to the data governance and patient privacy challenges present in multi-site clinical research, while also acknowledging the non-IID data heterogeneity and communication overhead that complicate federated implementation; and (3) it critically examines the emerging but clinically significant application of these integrative frameworks to outcome prediction in Fecal Microbiota Transplantation (FMT). This review combines mechanistic metabolic modeling, data-driven machine learning, and clinical feasibility considerations within a uniform analytical framework, offering a comprehensive and translational perspective relevant to the rational advancement of precision microbiome therapeutics.
2 Multi-omics integration for comprehensive microbiome characterization
Multi-omics integration combines supplementary data layers, including metagenomics (MG), meta-transcriptomics (MT), meta-proteomics (MP), and metabolomics (M), to surpass taxonomic profiling and achieve functional and mechanistic understanding of microbiome communities (). Each layer illustrates certain parameters. For instance, MG reveals genetic potential, MT indicates active gene expression, MP represents translated proteins, and M identifies end-products of microbial metabolism; yet their individual examination lacks essential interactions and emergent features (). Table 1 illustrates the multi-omics key pipelines and Tools, explaining their applications.
Table 1
| Pipeline/tool | Omics supported | Applications | Reference |
|---|---|---|---|
| gNOMO2 | ASV, MG, MT, MP | Wastewater, human gut; offering novel insights in both host-associated and free-living microbiome research. | () |
| MintTea | MG, MT, MP | Analyzing samples from a metabolic syndrome study, identifies a module associated with late-stage colorectal cancer | () |
| gNOMO | MG, MT, MP | Symbiosis studies, producing output visualizations, and analyzing multiple meta-omics data types. | () |
| PhyloPhlAn 3.0 | MG | Large-scale microbial genome characterization and phylogenetic analysis at multiple levels of resolution | () |
| QIIME 2 | 16S rRNA amplicon, ASV-based microbiome analysis | Microbiome diversity analysis, taxonomy classification, community profiling | () |
| MEGAN | MG, MT | Visual exploration and analysis of microbiome data | () |
| MetaPhlAn | MG | Profiling the composition of microbial communities from metagenomic shotgun sequencing | () |
| Mothur | 16S rRNA amplicon, OTU-based microbiome analysis | Sequence processing, taxonomic classification, OTU clustering | () |
Bioinformatics pipelines and applications.
ASV, Amplicon sequence variant; OTU: Operational Taxonomic Unit.
Multi-omics approaches have illuminated microbiome functions across diverse studies. In inflammatory bowel disease, integrated metagenomics, meta-transcriptomics, and metabolomics revealed facultative anaerobes, molecular disruptions in microbial transcription, metabolite pools and levels of antibodies in host serum (). In a multi-cohort application, MintTea identified disease-associated multi-omic modules combining microbial taxa and metabolites that collectively associate with disease phenotypes. It identified a module including Peptostreptococcus and Gemella species with fecal amino acids that progressively increase with cancer stage. This integrative approach demonstrated that capturing multi-layered co-variation can uncover therapeutic targets and biological mechanisms missed by single-omics analyses (). Within psoriasis patients, the researchers utilized a multi-omics approach to analyze the interactions between gut microbiota and metabolic pathways. By integrating microbiomes and metabolomic data, the study suggests that microbial dysbiosis may contribute to psoriasis pathogenesis by modulating lipid metabolism, inflammatory pathways, and oxidative stress responses (). Lastly, the comprehensive study by , aimed to elucidate the intricate interactions between the microbiota and host factors in the context of liver cirrhosis. This multi-omics profiling approach integrated gut microbiome data with plasma protein information, leading to the identification of specific classifiers associated with liver disease. The findings underscored that the microbiota-targeted biomarkers may be a powerful tool for diagnosis of different diseases (). Together, these studies illustrate both the potential contributions and scope of multi-omics approaches in advancing our understanding of complex diseases and identifying potential therapeutic targets (Figure 1).
Figure 1
3 Computational modeling approaches in microbiome systems biology
Computational modeling provides a quantitative framework to interpret multi-omics data and predict microbiome behavior under perturbations such as diet, drugs, or probiotics (). Unlike descriptive investigations, models represent host–microbe interactions, facilitating the simulation of metabolic flows, community dynamics, and host responses (). Within microbiome research, various significant modeling paradigms have emerged, varied in their assumptions, resolution, and suitability for therapeutic design ().
3.1 Flux balance analysis
Flux balance analysis (FBA) is a constraint-based mathematical framework used to estimate steady-state metabolic flux distributions in a metabolic network (). FBA has been used to model the metabolic phenotypes of individual gut microbes and to estimate the metabolic outputs of simplified microbial communities (). Additionally, it has been applied to investigate cross-feeding interactions in the gut microbiota. Community-scale extensions, such as OptCom and corresponding multi-species FBA frameworks, provide the explicit description of an interspecies metabolite exchange and the trade-offs between individual and community objectives. These approaches have been used to explore the effects of dietary perturbations, host-associated nutrient availability, and probiotic or therapeutic interventions on gut microbial metabolism ().
The primary drawbacks of FBA in gut microbiota research include its assumption of steady-state metabolism, reliance on the selected goal function and model constraints, and the increased complexity when applied to microbial communities. It must encompass metabolite interactions among several taxa and balance trade-offs between species-level and community-level growth (). These limitations, in gut microbiota, are amplified by interindividual dietary variation, missing metabolite measurements, strain-level diversity, and the difficulty of representing dynamic cross-feeding and host-microbe interactions in a single static model (). Additionally, the FBA community can be computationally and conceptually sensitive to how taxa are grouped, how exchange reactions are constrained, and how multi-species objectives are defined (; ). As a result, FBA is ideally regarded as a hypothesis-generating framework that is most effective when combined with metagenomics, metabolomics, transcriptomics, and experimental validation rather being employed as an independent predictive tool ().
3.2 ODE-based community models
Models based on ordinary differential equations (ODEs) highlight the temporal dynamics of microbial communities in relation to species interactions and environmental parameters. Generalized Lotka–Volterra (gLV) models provide a prevalent ODE framework whereby pairwise interaction coefficients indicate competitive, mutualistic, and predatory connections across microbial species (; ). These models have been utilized to investigate the temporal dynamics in gut microbiomes, covering the dynamics of antibiotic-associated Clostridioides difficile infection and subsequent recovery (). Dynamic variations in microbial communities are pivotal to human health and disease, and recognizing the interactions of microbial species and their environment might elucidate disease causes. Recent work has introduced the compositional Lotka–Volterra (cLV) model, a non-linear dynamical system that unifies generalized Lotka–Volterra (gLV) equations from community ecology with compositional data analysis techniques ().
Key challenges in gut microbiota studies include high parameter uncertainty due to limited longitudinal data, difficulty capturing higher-order interactions beyond paired terms, and challenges in linking relative abundance data to absolute microbial densities needed for meaningful growth-rate estimation (). Moreover, gLV and related ODE-based models frequently assume continuous, identically mixed populations, which may insufficiently describe spatial structure, strain-level variety, or host-mediated influences within the gut environment (). Consequently, these models are most efficient when integrated with experimental validation, absolute quantification techniques, and supplementary multi-omics data.
3.3 Agent-based modeling
Agent-based models (ABMs) simulate the actions of individual microbial agents. In contrast to ODE-based models, ABMs explicitly represent the spatial and temporal dynamics of the gut microbiota (). ABM characterizes dynamic systems as aggregates of distinct objects that engage with one another via established algorithmic protocols (; ). A fundamental advantage of ABM is its detailed focus; by establishing localized behavioral rules for each agent, researchers can construct resilient models even when the broader system dynamics are not completely understood (). Moreover, ABM facilitates real-time simulation of internal feedback loops and emergent behaviors, providing a robust framework for analyzing systemic complexity and evaluating their effects on specific agent attributes ().
Despite these advantages, ABMs face substantial challenges. They are computationally intensive, especially when scaled to the population sizes and diversity characteristic of real gut communities (). Parameterization remains challenging, as many agent-level rules require extensive sensitivity analysis and parameter estimation due to their highly integrated nature and substantial computational demands (). Moreover, model outcomes can be sensitive to rule definitions and initial conditions, making systematic calibration and uncertainty quantification challenging (; ). As a result, ABMs are most effective when used as exploratory or hypothesis-testing tools, complementing continuum models and experimental studies.
3.4 Genome-scale metabolic models
Genome-scale metabolic modeling (GEMs) is essential for reconstructing organism-specific metabolic networks from annotated genomes and use constraint-based methodologies (). GEMs have been developed for many gut commensals and pathogens, which can be integrated into community models that simulate metabolic cross-feeding, substrate competition, and the synthesis of essential metabolites such as short-chain fatty acids or bile acid derivatives (). These community-scale models can be integrated with host metabolic reconstructions to investigate how microbial metabolism alters host pathways, providing a mechanistic link between metagenomic potential and host phenotype (). When coupled with individualized food intake or metabolomics profiles, such models provide in silico testing of tailored interventions, including predicting which prebiotic fibers promote butyrate production or how probiotic strains influence amino acid or medication metabolism ().
Genome-scale metabolic models (GEMs) face several fundamental limitations when applied to microbiome studies. They rely on a steady-state flux assumption that intracellular metabolite concentrations remain constant, which fails to capture dynamic metabolic responses (; ). Standard GEMs typically lack explicit representation of gene regulation, transcription, translation, and signaling pathways that are essential for microbe–microbe and host–microbe communication (; ). Finally, the accuracy of GEMs is fundamentally limited by genomic knowledge gaps, where poorly annotated genes lead to dead-end metabolites and incomplete pathways that require manual, often speculative, gap-filling ().
3.5 Statistical and machine learning models
In addition to mechanistic models like GEMs and ODEs, statistical and machine learning models are widely used to elucidate high-dimensional relationships between microbiome characteristics and clinical symptoms, frequently employing regularized regression, random forests, or deep learning frameworks (). These models demonstrate proficiency in classification and risk prediction and can integrate multi-omics layers to enhance accuracy and resilience (). Nonetheless, their mechanistic interpretability is constrained, prompting hybrid methodologies whereby data-driven models highlight essential taxa or pathways that are subsequently integrated into the mechanistic frameworks mentioned above ().
Collectively, these modeling tools convert microbiome research from static representations into predictive, reproducible frameworks. In precision therapies, they facilitate the systematic design of interventions customized to individual microbiome configurations by predicting how specific alterations in nutrients, pharmaceuticals, or microbial compositions disseminate through metabolic and ecological networks. The subsequent sections will emphasize the integration of these models with multi-omics data and their implementation in workflows that transition from patient-specific measurements to actionable therapeutic suggestions.
4 Linking modeling and multi-omics for therapeutic design
The translational value of microbiome systems biology emerges when computational models are tightly integrated with multi-omics data to support precision therapeutic design through iterative, experimentally validated pipelines (). These frameworks typically combine patient-specific metagenomic, metabolomic, and host transcriptomic profiles to construct mechanistic or hybrid models that simulate microbial metabolism, host responses, and intervention outcomes (). Constraint-based reconstruction and analysis (COBRA) and physiologically based pharmacokinetic (PBPK) modeling enable quantitative prediction of microbiome–drug interactions, while community-scale genome-scale metabolic models (GEMs), constrained by gene expression or metabolite data, support personalized predictions of microbial function and therapeutic response (; ; ). Large-scale curated resources such as AGORA further allow integration of microbial and host metabolic networks to assess how dietary, probiotic, or pharmacological interventions alter bile acid metabolism, short-chain fatty acid production, and drug biotransformation in individual hosts ().
Experimental studies increasingly demonstrate the power of combining modeling with multi-omics for targeted microbiome modulation. Integrated meta-transcriptomic and metabolomic analyses in irritable bowel syndrome (IBS) patients have revealed functional disruptions not apparent from taxonomic data alone, highlighting new therapeutic targets (). In parallel, longitudinal multi-omics measurements in gnotobiotic mouse models exposed to bacteriophages have shown how dynamic modeling can capture direct and indirect community-wide effects of phage therapy, including metabolic rewiring and non-target taxa responses (). Data-driven approaches, including machine-learning-based optimization of multi-strain probiotic formulations, further illustrate how predictive models can guide the design of interventions with favorable metabolic outputs in complex gut environments (). Together, these studies underscore the potential of integrated modeling and multi-omics approaches to enable rational, mechanism-based microbiome therapies.
4.1 Clinical translation: FMT and microbiome therapeutics
Fecal microbiota transplantation (FMT) represents one of the most clinically advanced microbiome-based therapeutic strategies, with established efficacy in recurrent Clostridioides difficile infection () and ongoing clinical investigation in inflammatory bowel disease () and metabolic syndrome (). Machine learning models trained on donor-recipient microbiome composition have shown promising predictive performance for FMT clinical outcomes in C. difficile infection and ulcerative colitis cohorts (); however, the generalizability of these predictions across independent clinical populations remains to be established in prospective, adequately powered studies. Complementarily, genome-scale metabolic models (GEMs) have provided mechanistic insight into microbial cross-feeding interactions, metabolic capacity, and community-level engraftment dynamics following FMT (). The mechanistic principles underlying both approaches–responder stratification via supervised learning and metabolic compatibility modeling via GEMs–are directly relevant to the challenge of rational donor-recipient matching in FMT; however, no published study has yet directly integrated ML and GEM frameworks in an FMT-specific context. The development of such hybrid ML–GEM pipelines therefore represents a compelling and tractable direction for future investigation, with the potential to combine the predictive power of data-driven models with the mechanistic interpretability of constraint-based metabolic reconstruction (). Realizing this potential will require prospective cohort studies that collect paired metagenomic, metabolomic, and clinical outcome data at sufficient scale to train, validate, and externally test integrative models. Collectively, these advances highlight the potential of integrative modeling strategies to enable more rational stratification of microbiome-based therapeutic responses, suggesting a clearly defined and mechanistically grounded direction for future FMT research.
5 Limitations and challenges
Despite substantial advances, the integration of multi-omics data with computational modeling for microbiome-targeted therapeutics continues to face persistent methodological and translational challenges. A primary limitation arises from widespread data heterogeneity: microbiome studies differ markedly in sampling strategies, sequencing platforms, sequencing depth, and preprocessing pipelines, hindering cross-cohort comparability and limiting model generalizability across populations (; ). In addition, multi-omics datasets are often high-dimensional yet sparsely sampled, with incomplete or missing omics layers for many individuals. This imbalance can introduce biases in integrative analyses, reduce statistical power, and constrain model interpretability ().
Technical variation further complicates inference. Batch effects and laboratory-specific noise can obscure biologically meaningful signals, particularly when integrating datasets collected across institutions or over extended time periods (). On the modeling side, simplifying assumptions embedded in genome-scale metabolic models (GEMs), statistical learning frameworks, and dynamical simulations may fail to capture higher-order ecological interactions, spatial heterogeneity, or host-mediated effects that shape microbial community behavior in vivo (). Although many mechanistic and machine-learning models achieve strong performance within training cohorts, they frequently degrade when applied to external populations with distinct dietary patterns, geographic backgrounds, genetic predispositions, or treatment histories (). Furthermore, the increasing reliance on complex machine-learning architectures has raised concerns about model transparency and clinical interpretability, as “black-box” predictions based on latent features can be difficult for clinicians to justify in therapeutic decision-making ().
These challenges are further amplified in federated analytics settings, where data remain distributed across institutions rather than centrally pooled. In such frameworks, heterogeneity in sequencing platforms, pre-processing pipelines, and cohort composition cannot be fully resolved through global harmonization, leading to non-identically distributed (non-IID) data distributions that may bias model aggregation and degrade convergence across sites (; ). Moreover, federated analytics introduces additional system-level constraints, including increased inter-site communication overhead and synchronization latency due to repeated exchange of model parameters or gradients during training, which can be further exacerbated by uneven computational and networking capacities among participating centers in biomedical federations (). Addressing these limitations will require federated learning strategies that explicitly account for domain heterogeneity and communication efficiency to ensure robustness and scalability in large, multi-center microbiome studies.
6 Future perspectives
Advances in computational modeling and multi-omics integration are progressively shifting microbiome research from descriptive analysis toward predictive, individualized therapeutic development, although substantial methodological and clinical challenges remain before this potential can be fully realized. Among the most compelling near-term directions, microbiome digital twins – dynamic, patient-specific computational simulations that incorporate longitudinal multi-omics measurements, host genetics, and real-time environmental exposures – may constitute the next frontier in this field, potentially enabling virtual testing of interventions prior to clinical implementation (; ). These virtual models could simulate individual responses to dietary interventions, synbiotics, or microbiota-modulating medications, thereby supporting a shift from trial-and-error to mechanism-guided precision microbiome therapy.
Artificial intelligence (AI) will be pivotal in amplifying these methodologies. Deep learning architectures capable of learning complex representations from raw multi-omics data, combined with large language models trained on biomedical literature and clinical trial results, are expected to reveal new therapeutic hypotheses and facilitate the design of rational multi-strain microbial consortia (; ). Integration with wearable biosensors and continuous metabolomic monitoring platforms may provide closed-loop systems that adaptively modify therapies depending on real-time host–microbiome interactions.
Achieving clinical translation will require multidisciplinary collaborations to establish standardized platforms for model sharing, validation across diverse global populations, and regulatory frameworks for computational diagnostics (). Prospective randomized controlled trials that incorporate computational model outputs as co-primary endpoints, combined with real-world data from electronic health records and longitudinal microbiome biobanks, will be essential to generate the evidence base required for regulatory approval. While significant challenges in standardization, model generalizability, and clinical interpretability remain to be resolved, the continued convergence of mechanistic modeling, data-driven machine learning, and federated multi-institutional infrastructure represents a credible and increasingly actionable trajectory toward a comprehensive and clinically actionable understanding of the human gut microbiome.
7 Conclusion
This mini-review has examined the principal computational and multi-omics methodologies advancing precision gut microbiome therapeutics, with emphasis on their mechanistic underpinnings, analytical strengths, and inherent limitations. Flux balance analysis and genome-scale metabolic models (GEMs) provide constraint-based frameworks for modeling individual and community-level metabolic phenotypes, but are bounded by the steady-state assumption, the absence of explicit regulatory network representation, and genomic annotation gaps that require careful evaluation when interpreting predictive outputs. ODE-based community models, including generalized Lotka–Volterra variants, capture temporal microbiome dynamics but face challenges in parameter identifiability, representation of spatial structure, and linkage to absolute microbial abundances. Agent-based models offer granular, spatially explicit simulations at the cost of high computational demand and sensitivity to rule definitions and initial conditions. Across all paradigms, multi-omics integration with the combination of metagenomics, meta-transcriptomics, and metabolomics, is essential to connect computational predictions to experimentally verifiable, clinically relevant microbiome functions.
Emerging hybrid ML–GEM frameworks show early promise in improving phenotype predictions beyond what either approach achieves alone, although their prospective validation in clinical microbiome cohorts remains an open and important research priority. Federated analytics offers a pragmatic path toward training generalizable models on large, distributed microbiome datasets while preserving patient privacy, but its implementation is non-trivial: data heterogeneity across sequencing platforms and preprocessing pipelines, non-IID data distributions, and inter-site communication overhead are technical challenges requiring explicit methodological solutions. Realizing the translational potential of microbiome systems biology will further require standardized multi-omics protocols, explainable AI architectures transparent to clinical end-users, and collaborative regulatory frameworks for computational diagnostics. Taken together, the convergence of mechanistic modeling, data-driven machine learning, and multi-institutional federated infrastructure provides a rigorous and increasingly actionable foundation for precision microbiome medicine.
Statements
Author contributions
AD: Conceptualization, Data curation, Investigation, Resources, Visualization, Software, Writing – original draft, Writing – review & editing, Validation. MTM: Supervision, Project administration, Funding acquisition, Writing – original draft, Writing – review & editing, Validation.
Funding
The author(s) declared that financial support was not received for this work and/or its publication.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Correction note
This article has been corrected with minor changes. These changes do not impact the scientific content of the article.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
1
Aminian-DehkordiJ.MontazeriF.TamadonA.MofradM. R. K. (2025). Systems biology and microbiome innovations for personalized diabetic retinopathy management. NPJ Syst. Biol. Appl.11, 133. doi: 10.1038/s41540-025-00607-w. PMID:
2
AnG. (2008). Introduction of an agent-based multi-scale modular architecture for dynamic knowledge representation of acute inflammation. Theor. Biol. Med. Model.5, 11. doi: 10.1186/1742-4682-5-11. PMID:
3
AnG.MiQ.Dutta‐MoscatoJ.VodovotzY. (2009). Agent‐based models in translational systems biology. WIREs. Syst. Biol. Med.1, 159–171. doi: 10.1002/wsbm.45. PMID:
4
AnkrahN. Y. D.BernsteinD. B.BiggsM.CareyM.EngevikM.García-JiménezB.et al. (2021). Enhancing microbiome research through genome-scale metabolic modeling. mSystems6, 1–8. doi: 10.1128/mSystems.00599-21. PMID:
5
ArikanM.MuthT. (2024). gNOMO2: a comprehensive and modular pipeline for integrated multi-omics analyses of microbiomes. GigaScience13, giae038. doi: 10.1093/gigascience/giae038. PMID:
6
AsnicarF.ThomasA. M.BeghiniF.MengoniC.ManaraS.ManghiP.et al. (2020). Precise phylogenetic analysis of microbial isolates and genomes from metagenomes using PhyloPhlAn 3.0. Nat. Commun.11, 2500. doi: 10.1038/s41467-020-16366-7. PMID:
7
BauerE.ThieleI. (2018). From network analysis to functional metabolic modeling of the human gut microbiota. mSystems3, e00209-17. doi: 10.1128/mSystems.00209-17. PMID:
8
BiggsM. B.MedlockG. L.KollingG. L.PapinJ. A. (2015). Metabolic network modeling of microbial communities. WIREs. Syst. Biol. Med.7, 317–334. doi: 10.1002/wsbm.1308. PMID:
9
BolyenE.RideoutJ. R.DillonM. R.BokulichN. A.AbnetC. C.Al-GhalithG. A.et al. (2019). Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. Nat. Biotechnol.37, 852–857. doi: 10.1038/s41587-019-0209-9. PMID:
10
CaldeiraL. F.BorbaH. H.ToninF. S.WiensA.Fernandez-LlimosF.PontaroloR. (2020). Fecal microbiota transplantation in inflammatory bowel disease patients: a systematic review and meta-analysis. PloS One15, e0238910. doi: 10.1371/journal.pone.0238910. PMID:
11
CammarotaG.IaniroG.TilgH.Rajilić-StojanovićM.KumpP.SatokariR.et al. (2017). European consensus conference on faecal microbiota transplantation in clinical practice. Gut66, 569–580. doi: 10.1136/gutjnl-2016-313017. PMID:
12
ChoiG. J.KangH. (2025). Heterogeneity in meta-analyses: an unavoidable challenge worth exploring. Korean. J. Anesthesiol.78, 301–314. doi: 10.4097/kja.25001. PMID:
13
DayanI.RothH. R.ZhongA.HarouniA.GentiliA.AbidinA. Z.et al. (2021). Federated learning for predicting clinical outcomes in patients with COVID-19. Nat. Med.27, 1735–1743. doi: 10.1038/s41591-021-01506-3. PMID:
14
DienerC.GibbonsS. M.Resendis-AntonioO. (2020). MICOM: Metagenome-scale modeling to infer metabolic interactions in the gut microbiota. mSystems5, 10.1128/msystems.00606–19. doi: 10.1128/msystems.00606-19. PMID:
15
EetemadiA.RaiN.PereiraB. M. P.KimM.SchmitzH.TagkopoulosI. (2020). The computational diet: a review of computational methods across diet, microbiome, and health. Front. Microbiol.11, 393. doi: 10.3389/fmicb.2020.00393. PMID:
16
FaureL.MolletB.LiebermeisterW.FaulonJ.-L. (2023). A neural-mechanistic hybrid approach improving the predictive power of genome-scale metabolic models. Nat. Commun.14, 4669. doi: 10.1038/s41467-023-40380-0. PMID:
17
FerrocinoI.RantsiouK.McClureR.KosticT.de SouzaR. S. C.LangeL.et al. (2023). The need for an integrated multi-omics approach in microbiome science in the food system. Compr. Rev. Food Sci. Food Saf.22, 1082–1103. doi: 10.1111/1541-4337.13103. PMID:
18
FloresC.MillardS.SeekatzA. M. (2025). Bridging ecology and microbiomes: applying ecological theories in host-associated microbial ecosystems. Curr. Clin. Microbiol. Rep.12, 9. doi: 10.1007/s40588-025-00246-z. PMID:
19
FloresJ. E.ClaborneD. M.WellerZ. D.Webb-RobertsonB.-J. M.WatersK. M.BramerL. M. (2023). Missing data in multi-omics integration: recent advances through artificial intelligence. Front. Artif. Intell.6, 1098308. doi: 10.3389/frai.2023.1098308. PMID:
20
FreyE. (2010). Evolutionary game theory: theoretical concepts and applications to microbial communities. Phys. A. Stat. Mech. its. Appl.389, 4265–4298. doi: 10.1016/j.physa.2010.02.047. PMID:
21
García-JiménezB.Torres-BaceteJ.NogalesJ. (2021). Metabolic modelling approaches for describing and engineering microbial communities. Comput. Struct. Biotechnol. J.19, 226–246. doi: 10.1016/j.csbj.2020.12.003. PMID:
22
GibbonsS. M.DuvalletC.AlmE. J. (2018). Correcting for batch effects in case-control microbiome studies. PloS Comput. Biol.14, e1006102. doi: 10.1371/journal.pcbi.1006102. PMID:
23
GottsteinW.OlivierB. G.BruggemanF. J.TeusinkB. (2016). Constraint-based stoichiometric modelling from single organisms to microbial communities. J. R. Soc Interface13, 20160627. doi: 10.1098/rsif.2016.0627. PMID:
24
GreenblumS.ChiuH.-C.LevyR.CarrR.BorensteinE. (2013). Towards a predictive systems-level model of the human microbiome: progress, challenges, and opportunities. Curr. Opin. Biotechnol.24, 810–820. doi: 10.1016/j.copbio.2013.04.001. PMID:
25
GuC.KimG. B.KimW. J.KimH. U.LeeS. Y. (2019). Current status and applications of genome-scale metabolic models. Genome Biol.20, 121. doi: 10.1186/s13059-019-1730-3. PMID:
26
GuthrieL.KellyL. (2019). Bringing microbiome-drug interaction research into the clinic. EBioMedicine44, 708–715. doi: 10.1016/j.ebiom.2019.05.009. PMID:
27
HeinkenA.ThieleI. (2022). Microbiome modelling toolbox 2.0: efficient, tractable modelling of microbiome communities. Bioinformatics38, 2367–2368. doi: 10.1093/bioinformatics/btac082. PMID:
28
HsuB. B.GibsonT. E.YeliseyevV.LiuQ.LyonL.BryL.et al. (2019). Dynamic modulation of the gut microbiota and metabolome by bacteriophages in a mouse model. Cell Host Microbe25, 803–814.e5. doi: 10.1016/j.chom.2019.05.001. PMID:
29
HusonD. H.BeierS.FladeI.GórskaA.El-HadidiM.MitraS.et al. (2016). MEGAN community edition - interactive exploration and analysis of large-scale microbiome sequencing data. PloS Comput. Biol.12, e1004957. doi: 10.1371/journal.pcbi.1004957. PMID:
30
ImreyP. B. (2020). Limitations of meta-analyses of studies with high heterogeneity. JAMA Netw. Open3, e1919325. doi: 10.1001/jamanetworkopen.2019.19325. PMID:
31
JacobsJ. P.LagishettyV.HauerM. C.LabusJ. S.DongT. S.TomaR.et al. (2023). Multi-omics profiles of the intestinal microbiome in irritable bowel syndrome and its bowel habit subtypes. Microbiome11, 5. doi: 10.1186/s40168-022-01450-5. PMID:
32
JosephT. A.ShenhavL.XavierJ. B.HalperinE.Pe’erI. (2020). Compositional Lotka-Volterra describes microbial dynamics in the simplex. PloS Comput. Biol.16, e1007917. doi: 10.1371/journal.pcbi.1007917. PMID:
33
KantS.DeepikaRoyS. (2025). Integrative multi-omics and artificial intelligence: a new paradigm for systems biology. Omi. A. J. Integr. Biol.29, 576–587. doi: 10.1177/15578100251392371. PMID:
34
KootteR. S.LevinE.SalojärviJ.SmitsL. P.HartstraA. V.UdayappanS. D.et al. (2017). Improvement of insulin sensitivity after lean donor feces in metabolic syndrome is driven by baseline intestinal microbiota composition. Cell Metab.26, 611–619.e6. doi: 10.1016/j.cmet.2017.09.008. PMID:
35
KumarM.JiB.ZenglerK.NielsenJ. (2019). Modelling approaches for studying the microbiome. Nat. Microbiol.4, 1253–1267. doi: 10.1038/s41564-019-0491-9. PMID:
36
LangeE.KranertL.KrügerJ.BenndorfD.HeyerR. (2024). Microbiome modeling: a beginner’s guide. Front. Microbiol.15. doi: 10.3389/fmicb.2024.1368377. PMID:
37
LiT.SahuA. K.TalwalkarA.SmithV. (2020). Federated learning: challenges, methods, and future directions. IEEE Signal Process. Mag.37, 50–60. doi: 10.1109/MSP.2020.2975749. PMID:
38
LinC.CulverJ.WestonB.UnderhillE.GorkyJ.DhurjatiP. (2018). GutLogo: agent-based modeling framework to investigate spatial and temporal dynamics in the gut microbiome. PloS One13, e0207072. doi: 10.1371/journal.pone.0207072. PMID:
39
Lloyd-PriceJ.ArzeC.AnanthakrishnanA. N.SchirmerM.Avila-PachecoJ.PoonT. W.et al. (2019). Multi-omics of the gut microbial ecosystem in inflammatory bowel diseases. Nature569, 655–662. doi: 10.1038/s41586-019-1237-9. PMID:
40
MagnúsdóttirS.HeinkenA.KuttL.RavcheevD. A.BauerE.NoronhaA.et al. (2017). Generation of genome-scale metabolic reconstructions for 773 members of the human gut microbiota. Nat. Biotechnol.35, 81–89. doi: 10.1038/nbt.3703. PMID:
41
McCullochJ.GeJ.WardJ. A.HeppenstallA.PolhillJ. G.MallesonN. (2022). Calibrating agent-based models using uncertainty quantification methods. J. Artif. Soc Soc Simul.25. doi: 10.18564/jasss.4791
42
MedinaR. H.KutuzovaS.Hernández MedinaR.KutuzovaS.NielsenK. N.JohansenJ.et al. (2022). Machine learning and deep learning applications in microbiome research. ISME. Commun.2, 1–7. doi: 10.1038/s43705-022-00182-9. PMID:
43
Molina OrtizJ. P.McClureD. D.HolmesA.RiceS. A.ReadM. N.ShanahanE. R. (2025). Genome-scale metabolic modelling of human gut microbes to inform rational community design. Gut. Microbes17, 2534673. doi: 10.1080/19490976.2025.2534673. PMID:
44
MullerE.ShiryanI.BorensteinE. (2024). Multi-omic integration of microbiome data for identifying disease-associated modules. Nat. Commun.15, 2621. doi: 10.1038/s41467-024-46888-3. PMID:
45
Muñoz-BenaventM.HartkopfF.Van Den BosscheT.PiroV. C.García-FerrisC.LatorreA.et al. (2020). gNOMO: a multi-omics pipeline for integrated host and microbiome analysis of non-model organisms. NAR. Genomics Bioinforma.2, lqaa058. doi: 10.1093/nargab/lqaa058. PMID:
46
NagarajanK.NiC.LuT. (2022). Agent-based modeling of microbial communities. ACS Synth. Biol.11, 3564–3574. doi: 10.1021/acssynbio.2c00411. PMID:
47
OrthJ. D.ThieleI.PalssonB.ØCheckt. a. (2010). What is flux balance analysis? Nat. Biotechnol.28, 245–248. doi: 10.1038/nbt.1614. PMID:
48
PiresL.Gonzalez-ParamásA. M.HelenoS. A.CalhelhaR. C. (2024). Gut microbiota as an endocrine organ: unveiling its role in human physiology and health. Appl. Sci.14, 9383. doi: 10.3390/app14209383. PMID:
49
PleyerJ.FleckC. (2023). Agent-based models in cellular systems. Front. Phys.10. doi: 10.3389/fphy.2022.968409
50
PriceN. D.ReedJ. L.PalssonB.Ø. (2004). Genome-scale models of microbial cells: evaluating the consequences of constraints. Nat. Rev. Microbiol.2, 886–897. doi: 10.1038/nrmicro1023. PMID:
51
QinN.YangF.LiA.PriftiE.ChenY.ShaoL.et al. (2014). Alterations of the human gut microbiome in liver cirrhosis. Nature513, 59–64. doi: 10.1038/nature13568. PMID:
52
RiekeN.HancoxJ.LiW.MilletarìF.RothH. R.AlbarqouniS.et al. (2020). The future of digital health with federated learning. NPJ Digit. Med.3, 119. doi: 10.1038/s41746-020-00323-1. PMID:
53
Sabih Ur RehmanS.NasarM. I.MesquitaC. S.Al KhodorS.NotebaartR. A.OttS.et al. (2025). Integrative systems biology approaches for analyzing microbiome dysbiosis and species interactions. Brief. Bioinform.26, bbaf323. doi: 10.1093/bib/bbaf323. PMID:
54
SahuA.BlätkeM.-A.SzymańskiJ. J.TöpferN. (2021). Advances in flux balance analysis by integrating machine learning and mechanism-based models. Comput. Struct. Biotechnol. J.19, 4626–4640. doi: 10.1016/j.csbj.2021.08.004. PMID:
55
SatyaS.SharmaS.ChoudharyG.KaushikG. (2024). “ Advances in environmental microbiology: a multi-omic perspective BT - Microbial omics in environment and health,” in Microbial Omics in Environment and Health. Eds. KesheriM.KanchanS.SalisburyT. B.SinhaR. P. ( Springer Nature Singapore, Singapore), 175–204. doi: 10.1007/978-981-97-1769-9_7
56
SchlossP. D.WestcottS. L.RyabinT.HallJ. R.HartmannM.HollisterE. B.et al. (2009). Introducing mothur: open-source, platform-independent, community-supported software for describing and comparing microbial communities. Appl. Environ. Microbiol.75, 7537–7541. doi: 10.1128/AEM.01541-09. PMID:
57
SchroederW. L.SuthersP. F.WillisT. C.MooneyE. J.MaranasC. D. (2024). Current state, challenges, and opportunities in genome-scale resource allocation models: a mathematical perspective. Metabolites14. doi: 10.3390/metabo14070365. PMID:
58
SegataN.WaldronL.BallariniA.NarasimhanV.JoussonO.HuttenhowerC. (2012). Metagenomic microbial community profiling using unique clade-specific marker genes. Nat. Methods9, 811–814. doi: 10.1038/nmeth.2066. PMID:
59
SenP.OrešičM. (2023). Integrating omics data in genome-scale metabolic modeling: a methodological perspective for precision medicine. Metabolites13. doi: 10.3390/metabo13070855. PMID:
60
ShahidU. (2025). Microbiome-guided precision medicine: mechanistic insights, multi-omics integration, and translational horizons. J. Precis. Med. Heal. Dis.3, 100018. doi: 10.1016/j.premed.2025.100018. PMID:
61
ShtosselO.TurjemanS.RiuminA.GoldbergM. R.ElizurA.BekorY.et al. (2023). Recipient-independent, high-accuracy FMT-response prediction and optimization in mice and humans. Microbiome11, 181. doi: 10.1186/s40168-023-01623-w. PMID:
62
SteinR. R.BucciV.ToussaintN. C.BuffieC. G.RätschG.PamerE. G.et al. (2013). Ecological modeling from time-series inference: insight into dynamics and stability of intestinal microbiota. PloS Comput. Biol.9, e1003388. doi: 10.1371/journal.pcbi.1003388. PMID:
63
SuccurroA.MoejesF. W.EbenhöhO. (2017). A diverse community to study communities: integration of experiments and mathematical models to study microbial consortia. J. Bacteriol.199, 10.1128/jb.00865–16. doi: 10.1128/JB.00865-16. PMID:
64
SudhakarP.MachielsK.VerstocktB.KorcsmarosT.VermeireS. (2021). Computational biology and machine learning approaches to understand mechanistic microbiome-host interactions. Front. Microbiol.12. doi: 10.3389/fmicb.2021.618856. PMID:
65
TarazonaS.LopesM. B.KlammsteinerT.IbrahimiE.EckenbergerJ.TondaA.et al. (2023). Machine learning approaches in microbiome research: challenges and best practices14, 1–21. doi: 10.3389/fmicb.2023.1261889
66
TarziC.ZampieriG.SullivanN.AngioneC. (2024). Emerging methods for genome-scale metabolic modeling of microbial communities. Trends Endocrinol. Metab.35, 533–548. doi: 10.1016/j.tem.2024.02.018. PMID:
67
ten BroekeG.van VoornG.LigtenbergA. (2016). Which sensitivity analysis method should I use for my agent-based model? J. Artif. Soc Soc Simul.19, 5. doi: 10.18564/jasss.2857
68
ThieleI.PalssonB.Ø. (2010). A protocol for generating a high-quality genome-scale metabolic reconstruction. Nat. Protoc.5, 93–121. doi: 10.1038/nprot.2009.203. PMID:
69
WestfallS.CarracciF.EstillM.ZhaoD.WuQ.-L.ShenL.et al. (2021). Optimization of probiotic therapeutics using machine learning in an artificial human gastrointestinal tract. Sci. Rep.11, 1067. doi: 10.1038/s41598-020-79947-y. PMID:
70
WuF.JiangX.ChenG.ZhangL. (2025). Integrated microbiome and metabolome analysis reveals microbial-metabolic interactions in psoriasis pathogenesis. BMC Microbiol.25, 665. doi: 10.1186/s12866-025-04395-5. PMID:
71
YuL. C. (2018). Microbiota dysbiosis and barrier dysfunction in inflammatory bowel disease and colorectal cancers: exploring a common ground hypothesis. Journal of Biomedical Science. 25, 79. doi: 10.1186/s12929-018-0483-8
72
YuanH.ZhouJ.LiH.KangS.ParkS. (2025). A systems biology approach to memory health: integrating network pharmacology, gut microbiota, and multi-omics for health functional foods. Int. J. Mol. Sci.26, 6698. doi: 10.3390/ijms26146698. PMID:
73
ZomorrodiA. R.MaranasC. D. (2012). OptCom: a multi-level optimization framework for the metabolic modeling and analysis of microbial communities. PloS Comput. Biol.8, e1002363. doi: 10.1371/journal.pcbi.1002363. PMID:
Summary
Keywords
microbiome systems biology, multi-omics integration, genome-scale metabolic models, computational therapeutics, precision medicine, federated analytics, AI
Citation
Dewan A and Mascellino MT (2026) Computational and multi-omics systems biology for precision microbiome therapeutics. Front. Microbiomes 5:1842701. doi: 10.3389/frmbi.2026.1842701
Received
30 March 2026
Revised
23 April 2026
Accepted
29 April 2026
Published
15 May 2026
Corrected
21 May 2026
Volume
5 - 2026
Edited by
Sampathkumar Ranganathan, Sankara Nethralaya, India
Reviewed by
Sangavi Pandi, Karpagam Academy of Higher Education, India
Updates
Copyright
© 2026 Dewan and Mascellino.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Maria Teresa Mascellino, mariateresa.mascellino@uniroma1.it
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.