Translating Gut Microbiota into Diagnostics: A Multidimensional Approach for the Diagnosis of Inflammatory Bowel Disease

1) Evolution of 16S rRNA gene sequencing

16S rRNA gene sequencing has long been the standard approach for characterizing bacterial communities in fecal samples.¹⁵ Initial applications relied on Sanger sequencing, which, despite its high precision, had limited throughput. The emergence of second-generation sequencing (next-generation sequencing [NGS]) platforms such as 454 pyrosequencing, Ion Torrent, and Illumina MiSeq has enabled more efficient and cost-effective community profiling.¹⁶

Among these, 454 pyrosequencing was one of the earliest NGS platforms applied to microbiome research and provided longer reads (400 to 500 bp) for improved genus-level classification.^17,18 It contributed to early IBD studies by demonstrating characteristic dysbiosis in patients, most notably the depletion of Firmicutes and the enrichment of Proteobacteria.¹⁹ However, the platform was discontinued in 2016 due to high per-base costs, relatively low throughput, and the tendency to produce insertion–deletion errors in homopolymer regions. Thus, it now serves mainly as a historical reference and is unlikely to be integrated into future diagnostic applications.

Ion Torrent, introduced shortly after 454 pyrosequencing, presents an alternative approach based on semiconductor detection, enabling faster run times and lower reagent costs with read lengths comparable to those of its predecessor.²⁰ The platform provided read lengths similar to those of 454, offering faster running times and lower reagent costs. This platform has been applied in regional IBD studies. A study conducted in Russia used Ion Torrent sequencing to investigate microbial changes in patients with UC, reporting an increased abundance of pro-inflammatory genera, such as Haemophilus, Olsenella, and Butyrivibrio.²¹ In contrast, the relative abundance of several beneficial taxa, including the butyrate-producing genera Butyricimonas, Leuconostoc, Lactococcus, and Fusicatenibacter, was significantly reduced in patients with UC compared to that in healthy controls. However, the platform exhibits technical limitations, including reduced accuracy in homopolymer-rich sequences and batch variability, which limit its adoption in comparative studies requiring high precision.^22,23

The launch of the Illumina MiSeq platform represented a pivotal advancement in 16S rRNA sequencing. Using sequencing-by-synthesis with fluorescently labeled reversible terminators, MiSeq enables high-fidelity paired-end reads of up to 2×300 bp, allowing nearly full coverage of the V3–V4 region, which is currently the most widely targeted region in gut microbiome studies. The high accuracy, scalability, and cost-efficiency of MiSeq, combined with strong support from reference databases such as Greengenes2²⁴ and SILVA,²⁵ and widely adopted analysis pipelines such as QIIME2.²⁶ In a recent large-scale Korean study, Kim et al. used Illumina-based 16S rRNA sequencing to investigate microbial alterations in patients with IBD, identifying distinct taxonomic shifts between CD and UC.²⁷ Their analysis revealed enrichment of Escherichia-Shigella in CD, whereas genera such as Faecalibacterium, Bacteroides, and Dialister were significantly reduced in patients with UC. Furthermore, He et al.²⁸ characterized the gut microbiome in Chinese IBD cohorts and reported that microbial alpha diversity declined with increasing disease activity. Their analysis further revealed a depletion of Lachnospiraceae and Ruminococcaceae, accompanied by an enrichment of Escherichia-Shigella in IBD patients.

Third-generation sequencing technologies have recently extended the utility of 16S rRNA gene profiling by facilitating full-length sequencing. Platforms such as Pacific Biosciences (PacBio) Single Molecule Real-Time sequencing and Oxford Nanopore Technologies produce kilobase-length reads, surpassing region-specific resolution and allowing for the precise classification of species and even strains.^11,29 These capabilities are particularly relevant in IBD, where strain-level shifts and functionally divergent lineages within the same species may play important roles in disease heterogeneity. However, to date, no studies have directly applied full-length 16S rRNA long-read sequencing using PacBio or Nanopore in large-scale IBD cohorts. This represents the current gap in the field. Considering the increased resolution and gene-level coverage provided by long-read technologies, their integration into IBD microbiome workflows is expected to significantly advance biomarker discovery and provide a more accurate picture of dysbiosis at the functional and strain-specific levels. Although the cost and computational requirements remain to be considered, the integration of long-read sequencing into microbiome workflows represents a significant advancement toward more precise and informative microbial biomarker discovery.

Table 1 provides a structured comparison of widely used sequencing platforms, including second- and third-generation technologies. The table highlights their technical characteristics (read length, resolution, and cost), as well as their reported utilization in IBD studies. To facilitate clinical interpretation, we also indicate whether each platform has been validated in large-scale cohorts and its current clinical readiness for diagnostic applications.

2) Metagenome shotgun sequencing: microbial biomarker

Although 16S rRNA gene sequencing has served as the cornerstone for characterizing gut microbial communities, limitations in taxonomic resolution and a lack of functional information have prompted a growing shift toward shotgun metagenomic sequencing.³⁰ Unlike amplicon-based methods, which target specific regions of the bacterial 16S rRNA gene, shotgun metagenomics enables untargeted sequencing of the entire pool of microbial DNA within a sample. This comprehensive approach enables the simultaneous identification of bacteria, archaea, fungi, and viruses, thereby providing a broader view of the intestinal ecosystem.

One of the major advantages of metagenomic sequencing is its ability to characterize microbial communities at the species or strain level while simultaneously capturing the functional gene content.³¹ This high resolution is particularly valuable in the context of IBD, in which subtle microbial shifts and species/strain-level differences may have clinical relevance.^32,33 Moreover, metagenomic data can be used to identify genes involved in microbial metabolic capacity, antibiotic resistance, and virulence.³⁴ These features are increasingly being recognized as modulators of host–microbe interactions.³⁵

Several IBD cohort studies have used shotgun metagenomics to explore gut microbial signatures. Recently, Orejudo et al.³⁶ reported a comprehensive metagenomic profiling study that revealed consistent depletion of short-chain fatty acid (SCFA)–producing taxa, including Faecalibacterium prausnitzii and Roseburia, along with the enrichment of pro-inflammatory lineages, such as Escherichia coli, in newly diagnosed patients with IBD. These alterations were more pronounced during disease flares and were observed in patients naive to immunomodulatory therapy, revealing their potential role in disease initiation.³⁶ Furthermore, the study by Vich Vila et al.³² reported distinct microbiome signatures between patients with IBD and those with IBS, using high-resolution metagenomic sequencing of 1,792 individuals. Patients with IBD showed more pronounced dysbiosis than those with IBS, characterized by a reduced abundance and strain-level diversity of beneficial taxa, such as F. prausnitzii and Roseburia intestinalis, and increased levels of Bacteroides fragilis, Bacteroides vulgatus, and E. coli. In addition, a recent multicohort study by Ning et al.³⁵ applied metagenomic analysis across several international IBD cohorts, consistently identifying the depletion of anti-inflammatory commensals and the enrichment of pro-inflammatory strains, including Ruminococcus gnavus, B. fragilis, and E. coli. Although the study incorporated metabolomic data, microbiome-specific findings indicated that compositional and functional dysbiosis are conserved across geographically and clinically diverse patient populations.

Metagenomics has facilitated the development of microbial biomarker panels that are increasingly integrated with clinical and omics data for building diagnostic models. Despite these advantages, shotgun metagenomic sequencing presents technical challenges that must be carefully managed. High levels of host DNA contamination, especially in mucosal or low-biomass samples, can obscure microbial signals and reduce the effective sequencing depth.³⁷ Furthermore, accurate taxonomic classification depends on the completeness and uneven coverage of sequences, particularly in low-biomass samples.³⁸ However, with continued improvements in sequencing technologies, bioinformatics pipelines, and reference databases, shotgun metagenomics is emerging as a powerful tool for high-resolution, functionally beneficial microbiome studies in IBD.

3) Beyond the bacteria: virome and mycobiome signatures in IBD

The gut ecosystem harbors not only bacteria but also a diverse range of microorganisms, including eukaryotic viruses, prokaryotic viruses (bacteriophages), and fungi.^39,40 Advances in sequencing technologies have enabled systematic characterization of these communities, making it possible to detect taxa that exist at relatively lower abundance compared with bacteria and to uncover their potential contributions to intestinal homeostasis and disease. In the context of IBD, dysbiosis extends beyond bacterial taxa, and altered virome and mycobiome profiles are increasingly recognized as important disease-associated signatures.^41,42 These domains may provide additional layers of disease-specific biomarkers when analyzed in conjunction with bacterial communities.

Several studies have demonstrated that the gut virome is perturbed in IBD patients compared with healthy controls. Norman et al.⁴² performed metagenomic sequencing and reported increased diversity and richness of the enteric virome, with a notable expansion of Caudovirales bacteriophages in both CD and UC, suggesting that phage proliferation may represent a hallmark of IBD-associated dysbiosis. These findings indicate that virome signatures, particularly shifts in bacteriophage communities, may serve as non-bacterial biomarkers for IBD diagnosis and mechanistic studies.

Parallel research has highlighted the role of the gut mycobiome in IBD pathogenesis. Sokol et al.⁴¹ observed an increased abundance of Candida albicans and altered fungal diversity in CD, along with reduced levels of S. cerevisiae. These changes were associated with immune activation, including ASCA responses commonly used as a serological marker in CD. In another study, Hoarau et al. demonstrated that C. albicans strains isolated from IBD patients exhibited enhanced virulence and promoted pro-inflammatory cytokine production, further implicating the mycobiome in disease exacerbation.⁴³ Together, these studies suggest that fungal dysbiosis may complement bacterial markers in differentiating disease states.

Considering virome and mycobiome data alongside bacterial microbiome profiles enables a more holistic ecological analysis of gut dysbiosis in IBD. Network-based approaches that incorporate viral, fungal, and bacterial interactions may reveal key ecological drivers of disease. Importantly, such integrative analyses not only improve ecological interpretation but also provide a foundation for the systematic development of robust, multidimensional biomarkers that better capture the complexity of host–microbe dynamics in IBD.

1) DA analysis

Linear Discriminant Analysis Effect Size (LEfSe) is one of the earliest and most widely used tools for DA analysis.⁴⁹ LEfSe combines nonparametric statistical testing with linear discriminant analysis to identify microbial features that differ significantly between groups and have consistent effect sizes. It has been widely adopted owing to its interpretability and simplicity, especially in exploratory microbiome studies. Zhou et al.⁵⁰ applied LEfSe analysis to different intestinal sites and identified site-specific bacterial biomarkers associated with the IBD subtypes. Fusobacterium was enriched in CD, whereas Bifidobacteriaceae and Enterococcus were more abundant in UC. Furthermore, this study provided a detailed comparison of microbial signatures across lesion sites, revealing that microbial biomarkers vary according to anatomical location in both UC and CD, thereby highlighting the importance of spatial resolution in microbiome-based diagnostics. However, LEfSe does not adjust for covariates or the compositional structure of microbiome data, which limits its robustness in clinical applications.⁴⁵

To address these limitations, more advanced models such as Microbiome Multivariable Association with Linear Models (MaAsLin2) have been developed. MaAsLin2 was developed as a generalized linear modeling framework that accommodates both fixed and random effects, allowing for the adjustment of key confounding variables, such as age, medications, and sequencing batch.⁴⁸ MaAsLin3 further extends this functionality to multi-omics integration, supporting both continuous and categorical outcomes and handling relative or absolute abundance data.⁵¹ In a landmark study published in Nature Medicine, Zheng et al.⁵² applied MaAsLin2 to identify microbial features from metagenomic data that distinguished patients with IBD from healthy controls, yielding biomarkers such as R. gnavus, E. coli, and Clostridium innocuum. These were subsequently validated using ML classifiers, as discussed below. Despite its strengths, MaAsLin2 requires careful normalization and model specifications to avoid overfitting or false discoveries, particularly in high-dimensional or compositional datasets.^48,53

Analysis of Composition of Microbiomes with Bias Correction (ANCOM-BC) offers a robust alternative by directly addressing the compositional constraints inherent in microbiome data.⁵⁴ By estimating and correcting for sampling bias, ANCOM-BC produces bias-adjusted abundance estimates and provides false discovery rate (FDR)-controlled results. This method exhibits improved sensitivity in detecting differentially abundant features, particularly those present in low abundance. In a recent study using ANCOM-BC, 77 genera were identified as differentially abundant in patients with CD and 64 genera in patients with UC.⁵⁵ C. innocuum and Enterococcus were significantly enriched in both CD and UC, whereas Dialister and Fusobacterium showed higher abundances, specifically in CD. In contrast, Bifidobacterium was enriched in the UC group. Genera that were depleted in both IBD subtypes included Akkermansia and Roseburia. Faecalibacterium was significantly reduced only in patients with CD, whereas Bacteroides and Odoribacter were specifically decreased in patients with UC. These findings highlight both shared and distinct microbial signatures across IBD subtypes.

Despite its advantages, ANCOM-BC has limitations.⁵⁴ Although this effectively controls FDR and corrects for compositional bias, its computational complexity increases substantially with large datasets, particularly those containing thousands of features, making it challenging in resource-limited environments. Moreover, ANCOM-BC does not support flexible multivariate modeling for complex confounders in the same manner as tools such as MaAsLin2.⁵⁶ Although extended versions, such as ANCOM-BC2, have improved this limitation,⁵⁷ the need for method-aware applications remains critical.

Given the strengths and limitations of each DA approach, it is essential to select tools that align with the specific characteristics of the study design, dataset, and analytical objectives. Moreover, rather than relying on a single method, using multiple complementary DA tools and prioritizing microbial features that are consistently identified across analyses can enhance the robustness and generalizability of biomarker discovery. This multi-angle strategy mitigated method-specific bias and strengthened confidence in the clinical relevance of the resulting microbiome-based signatures.⁵⁸

2) Computational strategies for biomarker discovery (ML)

As microbiome datasets increase in size and complexity, traditional univariate approaches are insufficient for robust biomarker discovery. ML techniques have emerged as powerful alternatives for classification and feature selection, particularly in disease diagnosis.⁵⁹ These methods excel at modeling high-dimensional, nonlinear data and can integrate heterogeneous features, including microbial, metabolomic, and clinical variables.⁶⁰

Random forest (RF) classifiers are among the most frequently used ML algorithms for microbiome research. They built ensembles of decision trees to improve the classification performance and are often evaluated using metrics such as the area under the receiver operating characteristic (ROC) curve with area under the curve (AUC) value, specificity, and sensitivity.^35,61 In the context of IBD, several studies have successfully applied ML models trained on microbial profiles to distinguish patients from healthy individuals and differentiate UC from CD with high accuracy. For example, Halfvarson et al.⁶¹ used RF models to analyze microbiome and clinical data collected from patients with IBD over a 2-year period to assess subtype classification performance. The models were trained using operational taxonomic units (OTUs) abundances, clinical metadata (body mass index, sex, fecal calprotectin), and a unique “distance to healthy plane” metric derived from ordination space. The classifier achieved an overall prediction accuracy of 66.6% for distinguishing IBD subtypes from healthy controls, and the “distance to healthy plane” was identified as a key predictive feature. In another study, Manandhar et al.⁶² applied LEfSe analysis to identify significantly different microbial taxa between IBD and non-IBD samples using 16S rRNA gene sequencing data, followed by the implementation of various supervised ML algorithms, including RF, decision tree, elastic net, support vector machines with radial kernel, and neural networks, for disease classification. Among these, the RF model exhibited the best performance, achieving AUCs of >0.80 and consistently identifying microbial taxa of diagnostic relevance. Furthermore, to distinguish between CD and UC, LEfSe was used to select 117 bacterial taxa. When evaluated using a trained RF classifier, this feature set yielded high classification accuracy for CD and UC, with AUC values exceeding 0.91. In their multicohort study, Zheng et al.⁵² validated MaAsLin2-based microbial features using ML modeling to construct a noninvasive diagnostic classifier for IBD. The classifier showed strong performance in differentiating patients with UC and CD from non-IBD controls in the discovery cohort (AUCs >0.9). When applied to an independent validation cohort, the models maintained AUCs above 0.80 for both UC and CD versus non-IBD. In addition, when evaluated on a geographically and clinically diverse international multi-disease dataset, the models achieved AUCs of 0.7830 and 0.7206 for CD and UC, respectively, supporting their generalizability. These results highlight the clinical potential of combining statistical feature selection with the robust ML-based validation of microbiome-based diagnostics.

However, the performance and stability of ML-based biomarker discovery can be strongly influenced by cohort-specific effects and study heterogeneity, underscoring the need for cross-cohort integration strategies. The MMUPHin framework provides a statistical methodology for cross-cohort data integration and meta-analysis of microbiome studies.⁶³ By correcting for batch effects and study-specific heterogeneity, MMUPHin facilitates the identification of consistent biological patterns that are reproducible across independent datasets, thereby enhancing the reliability of microbiome-based biomarkers. Together with cross-cohort integration strategies such as MMUPHin, methodological rigor at the level of feature selection and validation remains equally important.

Although these studies revealed the utility of ML in microbiome-based diagnostics, many relied on a single DA method or ML-driven feature selection without comparative validation. As discussed above, integrating the results from multiple DA tools to identify robust biomarkers before ML modeling may enhance prediction accuracy and biological relevance. Moreover, unsupervised approaches such as microbial co-occurrence network analysis can highlight key hub taxa that may serve as important diagnostic features when validated through supervised learning models.⁶⁴ Thus, multistep pipelines combining DA, network inference, and ML represent promising directions for robust biomarker discovery and clinical translation in IBD (Fig. 1).

1) Metabolomic analysis

Metabolomics enables the profiling of small-molecule metabolites produced by both microbial and host metabolisms. In IBD, fecal and serum metabolomic analyses have revealed consistent alterations in pathways related to SCFAs, bile acids, tryptophan metabolism, and lipid signaling.⁶⁸ Reduced levels of SCFAs such as butyrate, which are known to support epithelial integrity and anti-inflammatory responses, have been frequently reported in patients with IBD. Similarly, the dysregulation of bile acid composition, particularly the ratio of primary to secondary bile acids, reflects disrupted microbial enzymatic activity and contributes to mucosal inflammation. These metabolites reflect microbial function as well as serve as potential functional biomarkers that may precede or parallel shifts in taxonomic composition. Their measurement provides a more immediate readout of microbiome activity and may be particularly useful for identifying subclinical disease states or monitoring therapeutic responses. Moreover, combining metabolomic data with metagenomic profiles enables the reconstruction of microbe–metabolite interaction networks that may be disease-specific.

Recent multicohort integration studies have used metabolomic data to enhance disease classification. For instance, Kim et al.⁶⁹ conducted untargeted and targeted serum metabolomics in a large Korean IBD cohort (n=346) and identified diagnostic panels capable of distinguishing patients with IBD from healthy controls and differentiating CD from UC. Their analysis revealed distinct alterations in tryptophan and bile acid metabolism, including elevated indole-3-acetic acid levels and reduced primary-to-secondary bile acid ratios. Using Boruta-based feature selection and ROC curve analysis, they reported high diagnostic accuracy (AUC >0.97 for CD and UC vs control, and AUC=0.714 for UC vs CD), highlighting the utility of serum metabolomic markers in both diagnosis and molecular subtyping of IBD. Ning et al.³⁵ showed that combining microbiome and metabolome features across cohorts improves the robustness of IBD-specific signatures and facilitates cross-study reproducibility. They reported that various organic acids and carnitine compounds were enriched in patients with IBD and that ML models based on these metabolomic features achieved high diagnostic performance in distinguishing patients with IBD from controls. These findings allowed the identification of robust diagnostic markers that generalized well across independent datasets. Integrating these features with metagenomic data enables the construction of microbe–metabolite interaction networks, thereby providing a mechanistic basis for identifying functionally relevant microbial biomarkers.

2) Transcriptomic analysis

Transcriptome analysis provides a dynamic snapshot of gene expression, allowing researchers to capture the active biological processes that occur in both the host and microbial components of the gut ecosystem. Metatranscriptomics, which involves sequencing total RNA from microbial communities, enables the identification of functionally active microbes and their expressed pathways.⁷⁰ For instance, metatranscriptomic analyses have revealed elevated microbial expression of pro-inflammatory genes related to oxidative stress responses, nutrient acquisition, and mucin degradation during active disease phases.^71,72 The integration of microbial and host transcriptomes enables the exploration of cross-kingdom regulatory dynamics, such as how microbiota-derived metabolites modulate host immune gene expression, thus providing a comprehensive understanding of the functional landscape of IBD pathogenesis and aiding in the development of composite molecular biomarkers.^73,74

3) Proteomic analysis

Although technically more demanding than other omics approaches, proteomics provides a direct measure of protein abundance and post-translational modifications, complementing transcriptomic insights and enabling the functional validation of candidate biomarkers.⁷⁵ Recent advancements in mass spectrometry, such as label-free quantification and Matrix-Assisted Laser Desorption/Ionization Time-of-Flight mass spectrometry, have significantly improved proteome depth and quantification accuracy in both host and microbial samples.⁷⁶ Host proteomic studies have revealed dysregulated proteins involved in mucosal healing, immune modulation, and extracellular matrix remodeling in IBD.^77-79

On the microbial side, proteomics, especially metaproteomics, has gained traction for linking microbial functions to disease phenotypes. Zhang et al.⁸⁰ performed metaproteomic profiling of pediatric IBD samples and revealed associations between microbiome composition and extracellular vesicle protein content in the gut, revealing that microbial proteins may influence host signaling pathways. However, they did not include a follow-up validation, highlighting the gap in translational utility.

Despite its promise, proteomics remains underutilized in microbiome-integrated IBD research compared with other omics layers. However, its role in validating candidate biomarkers and uncovering post-transcriptional regulation is increasingly being appreciated. Protein-level markers are particularly amenable to clinical translation, as they are more accessible for diagnostic development using antibody-based assays. A recent study by Lee et al.⁸¹ integrated proteomic, metagenomic, and clinical datasets to identify microbial signatures that predict responses to anti-tumor necrosis factor therapy. The authors showed that baseline differences in microbial protein expression profiles were associated with treatment outcomes, indicating that proteomic data could improve patient stratification and therapeutic decision-making for IBD.

As reviewed above, previous studies have used diverse sequencing platforms, biomarker analysis methods, and strategies to integrate multi-omics data. To provide a structured comparison, we summarize and categorize the key features of the representative studies in Table 2.^82-94 This comparative table enables the examination of the methodological strategies used in each study to identify robust and generalizable biomarkers, along with the validation efforts and the extent to which multi-omics approaches were used.