{
  "abstract": "Background Machine learning (ML) has been extensively applied in probiotic research for strain identification, fermentation optimisation, and computational disease-association prediction, predominantly in preclinical and in silico settings. However, the extent to which ML has been integrated into probiotic clinical research, where probiotics are directly administered or assessed in human subjects, remains poorly characterised. This study addresses that gap.Methods A systematic search was performed across PubMed, Scopus, and Web of Science databases according to PRISMA-ScR. Inclusion: human study design (RCT, observational, or case-control); probiotic application or intervention; ML/AI tool incorporated; original research article. Exclusion: non-human design; no probiotic in human participants; no ML tool; non-original article. A narrative synthesis of the findings was performed, grouping studies by clinical domain and ML function.Results Of 91 ML-probiotic-related studies identified for full-text analysis, 58 (64%) were non-clinical (preclinical, computational, or tool development) and excluded from the primary analysis, highlighting that most ML-probiotic studies remain at the discovery stage. A total of 33 clinical studies met all inclusion criteria. Publication accelerated sharply from 2021 onwards (88%), with 30% published in 2025 alone. Study designs included RCTs (n=16, 49%), observational/cohort studies (n=10, 30%), and other interventional designs (n=7, 21%). Random Forest (RF) was the most common algorithm (n=20, 61%), followed by XGBoost (n=7, 21%), SVM (n=6, 18%), and Logistic Regression (n=6, 18%). ML served four clinical roles: treatment response prediction and patient stratification (48%), biomarker discovery and classification (21%), microbiota profiling (15%), and outcome modelling (9%). Studies spanned 13 clinical domains ( IDDF2026-ABS-0469 Figure 1). Reported AUC ranged from 0.72 to 1.0, with RF predicting probiotic responders for constipation (AUC≤0.976), SIAMCAT LASSO separating probiotic from placebo microbiomes (AUC~0.99), XGBoost stratifying MASLD prebiotic responders (AUC=0.74–0.87), and SHAP identifying probiotic use as the second most important protective factor against neonatal feeding intolerance (XGBoost AUC=0.922).Conclusions While ML has been extensively applied in preclinical probiotic research, its integration into clinical research is emerging but promising. These 33 studies demonstrate that ML can deliver clinically actionable outputs, response prediction models, dysbiosis monitoring indices, and outcome classifiers, directly supporting personalised probiotic medicine. Prospective validation and standardised reporting of ML tools remain key priorities for future work.Abstract IDDF2026-ABS-0469 Figure 1",
  "authors": [
    {
      "affiliations": [
        "Microbiome Research Group, Research Centre for Life Science and Healthcare, Nottingham Ningbo China Beacons of Excellence Research and Innovation Institute (CBI), University of Nottingham Ningbo China, Ningbo, China"
      ],
      "name": "Loh Teng-Hern Tan"
    },
    {
      "affiliations": [
        "Pathogen Resistome, Virulome and Diagnostic Research Group (PathRiD), Jeffrey Cheah School of Medicine and Health Sciences, Monash University Malaysia, Bandar Sunway, Malaysia"
      ],
      "name": "Vengadesh Letchumanan"
    },
    {
      "affiliations": [
        "Microbiome Research Group, Research Centre for Life Science and Healthcare, Nottingham Ningbo China Beacons of Excellence Research and Innovation Institute (CBI), University of Nottingham Ningbo China, Ningbo, China"
      ],
      "name": "Jodi Woan-Fei Law"
    },
    {
      "affiliations": [
        "Microbiome Research Group, Research Centre for Life Science and Healthcare, Nottingham Ningbo China Beacons of Excellence Research and Innovation Institute (CBI), University of Nottingham Ningbo China, Ningbo, China"
      ],
      "name": "Learn-Han Lee"
    }
  ],
  "title": "IDDF2026-ABS-0469 From algorithm to intervention: the emerging role of machine learning in probiotic clinical research",
  "uid": "f4ed649e-c3b4-51b3-bcdb-41d4a6389b30"
}