Gut microbiome in gastrointestinal neoplasms: from mechanisms to precision therapeutic strategies

The GM drives tumorigenesis through two core mechanisms: indirect induction of host oxidative stress and direct secretion of genotoxic substances. The former primarily causes DNA damage, while the latter directly induces genomic instability, collectively constituting the molecular basis of GM-mediated carcinogenesis. Within direct genotoxic mechanisms, various pathogenic bacteria exhibit characteristic carcinogenic patterns. Salmonella enterica serovar Typhi ( S. Typhi)secretes typhoid toxin, which specifically induces DNA double-strand breaks and disrupts cell cycle regulation, a mechanism closely linked to GC development [14]. Biofilm Producing S. Typhi: Chronic Colonization and Development of Gallbladder Cancer [15]. This establishes a persistent inflammatory state, ultimately leading to DNA damage.

F. nucleatum has been shown to infiltrate CRC cells via its surface adhesin Fap2 protein. Its FadA protein binds to E-cadherin, triggering the Wnt/β-catenin/NF-κB signaling cascade and ultimately leading to genetic instability. FadA also contributes to tumor invasion through E-cadherin-mediated intercellular junctions [16, 17](Fig. 2A). In terms of mechanistic complexity, H. pylori is particularly illustrative: Its cytotoxin-associated gene A(CagA) protein-positive strains inject the oncoprotein CagA into gastric epithelial cells via a type IV secretion system (T4SS). CagA, through inhibiting the polarity-regulating kinase Partitioning-defective 1b (PAR1 b), disrupts tight junctions and microtubule dynamics stability, ultimately causing spindle assembly errors and chromosomal instability [4, 18]. Simultaneously, vacuolating cytotoxin A(VacA) toxin exacerbates genomic instability by interfering with autophagy homeostasis and inducing mitochondrial damage [9] (Table 1).

Bild vergrößern

In conclusion, these multi-level synergies profoundly explain the biological nature of GM driving tumor development by disrupting genomic integrity, from basic DNA damage to complex signal network dysregulation, jointly constituting the microbiota-mediated carcinogenic network.

Beyond direct effects, metabolites produced by GM significantly influence tumor progression by reshaping host metabolic pathways. One example is the role of short-chain fatty acids (SCFAs), such as butyrate, in regulating inflammation and maintaining the integrity of the intestinal barrier. Under normal physiological conditions, this is facilitated through G-protein coupled receptor 43 (GPR43) and Histone deacetylases (HDACs) signaling [19]. However, within the TME, SCFAs levels (e.g., butyrate/propionate) are significantly reduced. The depletion of these metabolites, which normally serve as the primary energy source for colonocytes, forces cells to shift towards glycolysis (the Warburg effect), providing ample ATP and biosynthetic precursors for cancer cell proliferation.

The deficiency of SCFAs further triggers cascading pathological effects: On one hand, reduced abundance of SCFAs-producing bacteria leads to dysbiosis, activating the Nucleotide-binding oligomerization domain-containing protein 2 (NOD2) receptor and triggering the NF-κB signaling axis. This promotes overexpression of pro-inflammatory cytokines like Interleukin-6(IL-6) and Tumor Necrosis Factor-α(TNF-α), establishing a chronic inflammatory microenvironment [20, 21].On the other hand, SCFAs concentrations are significantly reduced in young-onset colorectal cancer (yCRC), weakening their anti-inflammatory function [21]. Impaired mucosal barrier function intensifies intestinal permeability and weakens the innate immune defense ability. Notably, the systemic exhaustion of the SCFAs-producing microbiota has been observed in patients with yCRC, and this phenomenon is significantly positively correlated with the dysfunction of SCFAs synthesis, the loss of barrier integrity, and the persistent activation of inflammation [22, 23]. For instance, the abundance of SCFAs-producing bacteria such as Faecalibacterium prausnitzii, Blautia, and Roseburia in the CRC group (especially yCRC) was significantly reduced [24].

Disturbances in bile acid metabolism are equally relevant. Notably, an excess of deoxycholic acid (DCA) has been linked to the promotion of hepatocellular carcinoma (HCC), primarily through activation of the farnesol X receptor (FXR) and the G protein-coupled bile acid receptor (TGR5). These activations trigger oxidative stress and sustained inflammation in hepatocytes [25]. Additionally, Clostridium scindens-driven abnormalities in secondary bile acid metabolism have been shown to activate the Phosphatidylinositol 3-Kinase (PI3K)/Protein Kinase B (AKT) pathway, thereby stimulating the proliferation of CRC stem cells [26](Fig. 2B). The above-mentioned metabolic reprogramming processes jointly constitute the non-genotoxic network driven by the GM for tumor development.

The TME, as the hub of tumor progression, its formation is deeply dependent on the dynamic regulation of the GM, and one of the core drivers of its immunosuppressive state is the abnormal accumulation of lactic acid. The abnormal accumulation of lactic acid in the TME drives immune escape and constitutes the core pathological basis of tumor immunosuppression. An in-depth analysis of its regulatory network reveals that the GM mediates lactic acid enrichment through a dual-track pathway: On the one hand, Lactobacillus, Bifidobacterium, etc. directly produce lactic acid through dietary fiber fermentation, and the latter can permeate the TME through the circulatory system [27]; More crucially, the indirect regulatory axis, namely, dysbiosis (such as induction by RT and CT, promotes the abnormal proliferation of lactic acid-producing bacteria, and coincidentally leads to hyperglycolysis mediated by the "Wahlberg effect" of tumor cells, jointly resulting in significant acidification of the TME (pH 6.0–6.5). The acidic environment further inhibits the function of CD8⁺ T/NK cells, promotes the differentiation of regulatory T cells (Treg), weakens the response to immunotherapy [27], leads to tumor immune escape, and simultaneously reduces the response rate of immunotherapy.

The GM profoundly influences the TME through two primary pathways: direct activation of inflammatory pathways and indirect modulation of immune metabolism, acting as a “double-edged sword”. These microbes play central roles in maintaining immune homeostasis, regulating inflammatory signaling, and ensuring mucosal barrier integrity.

Specifically, certain pro-inflammatory members of the GM directly drive inflammatory responses. For instance, Hemolysin released by Aspergillus spp. triggers NOD-like receptor family pyrin domain containing 3 (NLRP3) inflammasome formation in monocytes, leading to Interleukin-1β(IL-1β) driven intestinal inflammation [3].Similarly, gut-colonizing Escherichia coli encoding the polyketide synthase genomic island(pks⁺ E. coli) secretes the genotoxin colibactin, releasing damage-associated molecular patterns (DAMPs) that activate the NF-κB pathway. This induces the production of pro-inflammatory cytokines Interleukin-6(IL-6), Interleukin-8(IL-8), and tumor Tumor necrosis factor-α(TNF-α), fostering a chronic inflammatory environment. This milieu favors the expansion of myeloid-derived suppressor cells (MDSCs) and suppresses T cell function, ultimately leading to immune tolerance and promoting tumorigenesis [13]. H. pylori can activate the NF-κB/IL-1β pathway through the synergistic effect of its bacterial components (such as peptidoglycan/ADP-heptaglucose) and CagA, forming a pro-cancer microenvironment and enhance the carcinogenic effect of CagA through a positive feedback loop [4].

On the other hand, probiotics such as Bifidobacterium have demonstrated significant immunomodulatory and mucosal repair functions, mainly by regulating immune responses and inhibiting inflammatory responses [28]. In the gut and lungs, Bifidobacterium promote mucosal healing through multiple mechanisms: they modulate adaptive immune responses, reducing the release of pro-inflammatory cytokines (e.g., IL-17 and IL-6) [29]. For example, Bifidobacterium animalis subsp. Lactis has been proven to have the effect of inhibiting IL-17, which is crucial for alleviating chronic inflammation (such as IBD) and promoting mucosal healing. Meanwhile, Bifidobacteria can enhance the intestinal barrier function (such as strengthening tight junctions), promote the regeneration of mucosal cells, and help maintain the balance of the microbiome, preventing the excessive growth of pathogens, thereby effectively repairing damaged mucosa [30].

Among the various beneficial effects of its regulation of the key pro-inflammatory cytokine IL-6 is particularly critical and closely linked to the efficacy of tumor immunotherapy. Its mechanisms include secreting anti-IL-17 substances, blocking the positive feedback loop between IL-6 and IL-17, and reducing the differentiation of T helper 17 (Th17) cells. Furthermore, by enhancing intestinal tight junctions to reduce endotoxin translocation into the bloodstream, Bifidobacterium can effectively suppress systemic inflammatory responses, including IL-6 release.[16].To sum up, Bifidobacterium can restore microecological balance by regulating the composition of the intestinal flora (such as increasing the ratio of bifidobacterium/E. coli), inhibiting key pro-inflammatory pathways (such as Th17/IL-17), and reducing the levels of core inflammatory factors (such as IL-6).

Notably, GM metabolites also represent an important indirect pathway influencing immune status. For example, lipopolysaccharide (LPS) leads to insufficient activation of NF-κB by specifically inhibiting the Toll-like receptor 4 (TLR4)/Myeloid differentiation factor 2 (MD-2) receptor. The result is a decrease in the secretion of pro-inflammatory factors and an increase in the secretion of immunosuppressive factors, which may eventually lead to tumor immunosuppression and tumor cell proliferation [31] (This effect is particularly worthy of attention in the context of ICIs treatment).

In the gastrointestinal tract, microbiota-derived metabolites support indirect immune regulation. By influencing cytokine networks, antigen-presenting cell activation, and lymphocyte differentiation, these compounds orchestrate a broad array of immune responses. One extensively studied mechanism involves Treg-mediated immune tolerance. In the gastric environment, Foxp3 + Tregs release tumor necrosis factor-β(TGF-β) and Interleukin-10(IL-10), suppressing CD8 + T cell activity and enabling tumor immune evasion [32]. Tryptophan-derived metabolites, including indole analogs, activate the aromatic hydrocarbon receptor (AhR). This activation induces the expression of indoleamine 2,3-dioxygenase (IDO1), drives the conversion of tryptophan to kynurenine, suppresses CD8 + T cell function [33]. Clinical data indicate a strong link between programmed cell death protein 1 (PD-1) inhibitor resistance and elevated AhR activity in the gut of CRC patients [34] (Fig. 2C).

The intestinal microbiota can not only prevent tumorigenesis and alleviate colitis or radiation enteritis associated with ICIs, but also may enhance the efficacy of immunotherapy by improving the immune microenvironment (such as enhancing the activity of CD8 + T cells), laying an important microbiological foundation for its application in tumor immunotherapy.

Nonetheless, using GM as a biomarker presents several challenges, including the low abundance of microbial populations in tumor tissues and the risk of sample contamination [43]. As a result, there remains no standardized approach for integrating macrogenomic and metabolomic datasets. Additional multicenter cohort studies are necessary to validate biomarker thresholds and ensure consistency [24].

Although GM has potential in cancer prevention and treatment, major obstacles still hinder its clinical application. A major issue is the absence of standardized methodologies in GM research; assay variability leads to inconsistent data, and the lack of agreement on biomarker thresholds, in combination with regulatory delays and limited safety data, slows the progress toward clinical implementation.

FMT is currently a very important means for the precise treatment of GI neoplasms. It assists tumor treatment by reshaping the ecological balance of the intestinal tract. Notably, FMT has become a key approach to addressing ICIs resistance in tumor immunotherapy [52]. Recent studies have focused on identifying ICIs responders and characterizing their potential as FMT donors. In a metastatic melanoma model, FMT from responders to anti-PD-1 therapy was associated with an improved objective response rate (ORR). This effect has been linked to the enrichment of Prevotella merdae, which promotes anti-tumor immunity through the activation of CXCR3 + CD8 + T cells. Moreover, responder-derived FMT has been shown to reduce pro-inflammatory mediators such as TNF-α and IL-1β, restore GM balance, and significantly lower the risk of immune-related adverse events (irAEs) [53]. The therapeutic value of healthy donors has also been highlighted. FMT from healthy individuals supports intestinal barrier integrity after RT by sustaining butyrate-producing bacteria like Faecalibacterium prausnitzii, maintaining SCFAs levels, and preserving specific pathogen-free (SPF) conditions. Notably, SPF mice subjected to RT displayed improved intestinal barrier function and survival rates rising to 75% [54]. Studies have confirmed that FMT of different donor types has differentiated characteristics, and while they bring clinical benefits, they still face many challenges. ICIs responders can induce targeted anti-tumor immune responses, yet may also carry bacterial strains associated with therapeutic resistance. Conversely, healthy donors present a lower-risk profile, though they may lack the specific capacity to modify the TME [55]. However, at present, for the FMT treatment of cancer patients, more effective donor types (ICIs responders vs. Healthy donors (FMT transplantation), there is still a lack of clear clinical trial evidence and literature conclusions, and there is significant controversy. To address this issue, many clinical trials and preclinical trials are currently available to further explore the effectiveness of FMT.

In addition to the related issues of donor population selection, its clinical efficacy also highly depends on strict donor screening criteria. Based on the current Gut 2017 and 2019 conconsensus [56, 57], a complete screening process should include preliminary questionnaires, detailed tests and dynamic monitoring to ensure the safety of donations, and be supplemented by standardized biobank management in line with international consensus guidelines, thereby providing safe and reliable FMT materials for clinical applications(Fig. 3).

Bild vergrößern

The core of engineered bacteria in treating cancer lies in achieving the targeted delivery of immunomodulatory compounds through engineered microorganisms. Existing clinical studies provide strong evidence for standardized microbial intervention: Three weeks after oral administration of Bifidobacterium BB-12, the serum IL-6 level of the patients decreased significantly by 81.5% (6.2 pg/mL vs 33.6 pg/mL in the control group, p < 0.001), the mortality rate decreased from 25 to 5%, and the hospital stay was shortened by 43%. And the improvement rate of CT imaging increased to five times that of the control group (70% vs 13.6%) [29].In addition, the current preclinical experiments include: In the field of metabolic engineering, the modified strains have successfully alleviated the toxicity of CT- for example, by selectively blocking the reactivation of SN-38 in the intestine, the diarrhea caused by irinotecan has been reducing the incidence by approximately 80% without affecting the microbial balance; Engineered butyrice-producing bacteria (such as Clostridium butyricum) can increase the concentration of SCFAs and effectively improve chemoid-induced intestinal mucolitis [26]. However, all of the above are still at the preclinical stage, and further clinical development is still needed.

Bacteriophages enhance immune regulation through three synergistic pathways: (1) Direct immune modulation by penetrating the blood–brain barrier to influence host physiology [58]; (2) Microbiota restructuring through lysing specific bacteria, thereby regulating metabolic networks and altering key co-metabolites (e.g., SCFAs and bile acids); (3) Co-evolutionary adaptation that leverages microenvironmental factors (nutrients, spatial structure) to sustain gut homeostasisn [59]. Their exceptional specificity originates from precise recognition of bacterial surface receptors, enabling targeted pathogen elimination without harming beneficial microbiota or eukaryotic cells.

Based on the above progress, combining FMT with engineered bacteria is expected to produce dual effects: “ecological restoration + targeted immune activation”[26]. Once FMT re-establishes GM diversity, co-administered engineered bacteria can deliver immunoregulatory agents such as PD-1 nano-antibodies or IL-15, further strengthening the anti-tumor response [52].

Preclinical models and early-phase clinical trials have observed benefits of the aforementioned engineered bacteria and phage therapies for GI neoplasms. However, their clinical application requires substantial further development. This strategy offers a promising, actively investigated approach to reversing cancer treatment resistance. It shows particular intervention potential in refractory patient populations. Nevertheless, significant challenges remain for standardizing FMT and implementing engineered bacterial therapies. Current findings must still be validated through large-scale, multicenter clinical trials. This validation is essential to confirm clinical translation value and establish evidence-based intervention guidelines. To make clinical translation viable, three core issues must be addressed: mechanistic understanding, donor selection, and the safety of engineered strains. Future research should adopt multi-omics approaches to design personalized therapeutic strategies and support the broader application of “microbial precision medicine” in oncology.

RT, as a conventional method for tumor treatment, can directly kill tumor cells by inducing DNA damage. However, recent studies have found that it may indirectly promote tumor progression by reshaping the TME. The core mechanism lies in the immunosuppressive transformation induced by RT: RT significantly upregulates the expression of Amphiregulin (AREG) in tumor cells by activating the type I interferon signaling pathway, thereby triggering the “badscopal effect”—a pathological process that drives distant metastasis [60, 61]. It is worth noting that off-target damage to the intestinal barrier caused by RT often leads to radiation enteritis, which not only limits the intensity of treatment but also seriously reduces the quality of life of patients. Chemotherapy-induced gastrointestinal (GI) toxicity is common with many chemo regimens. About 40% of patients receiving standard-dose chemo develop GI toxicity. This rate increases markedly with high-dose chemo, reaching about 60% to 100%[8]. The latest evidence suggests that the GM plays a crucial protective role in mitigating RT toxicity by modulating the host’s immune and oxidative stress responses [54].

The radioprotective mechanisms of GM operate through two primary pathways. Mechanistically, microbial metabolites mediate intestinal barrier repair: metabolites such as indole-3-carboxaldehyde (I3A) and kynurenic acid (KYNA) suppress NF-κB pathway activation, reducing intestinal IL-1β levels by 67% (P < 0.001) [45]. Concurrently, they enhance tight junction protein (e.g., occludin) expression, thereby mitigating systemic inflammation triggered by LPS translocation.

Furthermore, FMT demonstrates synergistic radioprotection by enriching SCFAs-producing bacteria (e.g., Roseburia spp.). This process boosts intestinal epithelial antioxidant capacity (2.1-fold increase in SOD activity) and reduces clinical radiation enteritis incidence by 42% (HR = 0.58) [54].

Notably, microbial dysbiosis may exacerbate RT induced metastatic risks. Clinical studies reveal that CRC patients receiving nCRT with enriched Fusobacterium and Enterococcus exhibit significantly lower pathological complete response (pCR) rates (OR = 3.24, 95% CI: 1.87–5.61) [62]. This effect correlates with microbiota-driven upregulation of the AREG/STAT3 signaling pathway.

The current clinical urgency stems from the irreversible predicament of radiotherapy toxicity management: Traditional symptomatic treatment is unable to block the metastatic cascade reaction mediated by microbiota disorder and LPS translocation. The individualized intervention strategy based on metabolomics shows breakthrough potential—the risk of radiation enteritis can be predicted through preoperative microbiota function analysis (such as I3A/KYNA level detection) (AUC = 0.82) [39]. The integration of the "microbiota—metabolites—immunity" regulatory network is becoming a new paradigm to solve the paradox of RT promoting metastasis and providing a precise solution for balancing tumor control and therapeutic toxicity.

The GM plays a core role in reducing CT toxicity by regulating the host's immune and metabolic pathways, and its core mechanism focuses on the drug reactivation process mediated by microbial enzymes. Take irinotikang as an example. After the drug is converted into the active metabolite SN-38 in the liver, it is acidified with gluconaldehyde to form the non-toxic SN-38G and is excreted into the intestine. The β-glucuronidase (GUS) secreted by Clostridium spp. Can hydrolyze SN-38G, re-release the toxic SN-38 molecule, directly damage the intestinal epithelial barrier, and induce severe diarrhea (incidence > 80%) [46]. Targeting this key toxic pathway, the GUS inhibition strategy shows clinical potential: New type of sterilization GUS inhibitors (such as traditional Chinese medicine (TCM) derivatives to hydrogen iso-eugenol) through the selective inhibition of bacteria Loop—type 1 GUS and enzyme activity, and an 80% lower incidence of diarrhea in preclinical models, at the same time avoid dysbacteriosis broad-spectrum antibiotics [63].

The GM exerts a clinically significant dual regulatory role on CT efficacy. On one hand, microbial metabolites such as butyrate enhance CT response by inducing p21 expression to arrest tumor cells in the G1 phase, thereby sensitizing them to 5-fluorouracil (5-FU)-mediated cytotoxicity. Concurrently, butyrate directly promotes tumor cell apoptosis via suppression of the AP-1 signaling pathway, synergistically augmenting 5-FU-induced cell death [64]. Conversely, F. nucleatum compromises CT efficacy through activation of the TLR4/MyD88 pathway. This cascade upregulates core autophagy machinery in tumor cells, enabling drug clearance and suppression of apoptosis. Critically, this mechanism confers chemoresistance, exemplified by a 3.1-fold reduction in oxaliplatin sensitivity in CRC models [26](Fig. 4).

Bild vergrößern

The GM, as the core regulator of the efficacy and toxicity of RT and CT, plays a “double-edged sword” role in the treatment window by dynamically shaping the host's immune response and metabolic pathways. In the context of RT, microbiota imbalance can amplify the bad effect. While in the field of CT, the key contradiction is concentrated on the drug toxicity mediated by microbial enzymes, which significantly limits the intensity of treatment.

The GM, as a key controller for “enhancing efficacy and reducing toxicity” in RT and CT, is a complex network that urgently needs to be deconstructed through multi-omics technology. With the development of synthetic biology, engineered colonizing bacterial communities (such as probiotics expressing GUS inhibitory peptides) are expected to achieve local toxicity blocking. The artificial intelligence-driven microbiota atlas will guide the design of individualized RT and CT regimens, ultimately promoting a paradigm shift in tumor treatment from "maximum tolerated dose" to "microbiome optimized treatment window". This transformation process not only requires interdisciplinary collaboration to crack the microbiota-host interaction mechanism, but also needs prospective clinical trials to verify the improvement benefits of microbial intervention on long-term survival.

The integration of spatial transcriptomics, metabolomics and in situ imaging of microorganisms provides a high-precision analytical tool for the three-dimensional interaction network of microbiota—immune cells—tumor cells in the TME [74]. For instance, the Meta-Spatial platform, by visualizing the co-localization phenomenon of F. nucleatum and M2 macrophages in CRC, revealed for the first time the mechanism by which the two mediate immune escape through the TLR4/NF-κB pathway [75].

It is notable that this technology further realizes the spatial distribution and localization of oncogenic bacteria at the forefront of tumor invasion [76], thereby identifying the key interaction hotspots that affect tumor metastasis and treatment resistance [77]. Combined with metabolomics analysis, researchers can visualize the concentration gradient distribution of bacterial metabolites (such as butyrates and genotoxins) in the TME and clarify their dynamic regulatory effects on the gene expression of adjacent cells. This technical system provides a spatial dynamic basis for individualized treatments such as probiotic intervention and FMT.

By integrating immune cells, vascular endothelium, and fluid microenvironment, the establishment of a high-fidelity bionic "immune-tumor-microbiota" organoid model can effectively simulate the dynamic complexity of the TME in vivo [78]. The integrated use of patient-derived tumor organoids (PDTOs) and autologous fecal microbiota establishes a personalized therapeutic prediction platform. This system not only evaluates treatment responses and toxicities(including prognostic outcomes in specific cohorts such as CRC patients with liver metastases)but also facilitates the screening of microbiota-targeted interventions, such as probiotic combinations and synthetic microbial consortia [79]. An experimental study confirmed that metronidazole reversibly inhibits F. nucleatum mediated FadA toxin of Wnt/β-catenin pathway abnormal activation [17]. At the same time, the model for targeted gene engineering of bacteria and safety testing provides a new way. However, maintaining the diversity of the microbiota, simulating physiological environments such as the intestinal hypoxia gradient and mucus layer, and standardizing the model construction process remain the core challenges currently faced.

The longitudinal integration of metagenomic, metabolomic, and single-cell transcriptomic data can analyze the causal association between the dynamic changes of the microbiota and the therapeutic response. Typical cases include the study of the mechanism by which the enrichment of Bacteroides leads to the decrease of SCFAs levels and thereby induces the exhaustion of CD8⁺ T cells [80]. For instance, Sultana A and colleagues recently employed an integrated multi-omics approach to analyze existing research. Their goal was to clarify the link between GM, their metabolites, and depression. The study revealed specific changes in the GM composition and metabolic profiles of depression patients [81]. These findings provide a basis for developing personalized precision treatment strategies. This work not only demonstrates the feasibility of integrated multi-omics analysis in clinical research but also helps uncover the interaction mechanisms between the disease and GM. It thereby lays the groundwork for innovative precision treatments.

Building on this foundation, the integration of multi-tiered data—encompassing spatial omics, organoid models, and clinical cohorts—enables the construction of a unified predictive model bridging molecular mechanisms to clinical phenotypes. This comprehensive framework establishes a theoretical basis for developing combinatorial biomarkers for precision diagnosis and prognosis. Central to this endeavor is overcoming critical bottlenecks in standardizing heterogeneous big-data integration, advancing multimodal fusion algorithms, and establishing cross-platform repositories to support systematic, large-scale discovery research.

The host-microbiota-drug interaction network constructed based on deep learning is accelerating the exploration of new therapeutic value for marketed drugs. It can mainly be predicted through the following three aspects: Drug metabolism prediction: Using the microbiota functional profile to evaluate the bioavailability and toxic metabolite generation of oral drugs. The mechanism of reusing old drugs: Through algorithms, new treatment directions are developed for drugs whose other therapeutic effects have been clearly defined. For example, Reveal rapamycin etRepMicro algorithm by inhibiting the mTOR pathway of the nuclear spindle coli sensitization chemotherapy [82], whereas metformin by activating Akkerrmansia AMPK pathway inhibition of CRC progression [83]; Toxicity risk early warning: Integrating microbiota-host genome data to predict adverse reactions in the immunotherapy phase. At present, technical challenges impede the combined use of AI therapies. These include the lack of standardized microbiome data, poor model stability, and unexplainable algorithms, all limiting AI's application in clinical translation. In the next research direction of precision treatment, we can further focus on microbiota-related pathways such as TLR signaling and bile acid metabolism, and systematically reposition the anti-tumor potential of existing drugs.

Artificial intelligence is deeply empowering the innovative development and personalized application of microbiota therapy: for instance, engineered bacteria design: predicting the colonization efficiency and safety of genetically engineered bacteria (such as tumor-targeted drug-producing bacteria) in organoid models; Customized intervention strategies: Recommend probiotic combinations based on multi-omics characteristics, or screen FMT donors through the microbiota-immune interaction network, and generate precise dietary intervention plans to regulate key metabolites (such as butyrate) [84]; Cutting-edge research demonstrates that machine learning can predict personalized changes in GM and their metabolites. For instance, PH-responsive engineered bacterial capsules can be designed through AI models to achieve targeted release of TLR ligands/cytokines in the intestinal tract, enhancing vaccine efficacy and reducing systemic toxicity [85]. By tailoring immune responses in the host, this approach may enhance vaccine efficacy [86]. These studies validate the clinical feasibility of using machine learning predictions to modulate host immunity for cancer treatment. This progress supports the development of personalized disease prevention strategies.

However, to date, the long-term biosafety verification of engineered bacteria and the standardized implementation of individualized interventions remain challenges to be addressed in clinical transformation.