Artificial intelligence is widely proposed against antimicrobial resistance, but how much of that literature could support clinical deployment is unclear. We searched PubMed (2016–2026) across four domains — genomic prediction, clinical risk models, therapy, and rapid diagnostics — retrieving all 2,446 records. Of 725 eligible studies, 80 were appraised at full text against the artificial-intelligence extensions of the standard risk-of-bias and reporting tools (PROBAST+AI, TRIPOD+AI). Selection favoured externally validated work, so the proportions are upper bounds. Output grew from 10 studies in 2016 to 142 in 2025, with no evidence that genuinely external evaluation became more likely (odds ratio 1.08/year, 95% CI 0.89–1.34). Of the 80, 45% were externally validated, 24% released a model a reader could apply, 23% assessed calibration, 16% justified a sample size, and five compared performance across patient characteristics against eight across sites; 9% called a random split of their own data external validation. Risk of bias was low in 16, unclear in 55 and high in 9, and none of the 18 deep-learning studies reported calibration (95% CI 0–17.6%). Validation, calibration and transparency limit this field, not accuracy. We derive the design constraints these impose on a culture-independent diagnostic platform, a projection with no performance data.

Artificial intelligence is widely proposed against antimicrobial resistance, but how much of that literature could support clinical deployment is unclear. We searched PubMed (2016–2026) across four domains, retrieving all 2446 records. Of 725 eligible studies, 80 were appraised at full text against the artificial-intelligence appraisal extensions (PROBAST+AI, TRIPOD+AI). Selection favoured externally validated work, so the validation figures are upper bounds. Output grew from 10 studies in 2016 to 142 in 2025; within the appraised set there was no evidence that genuinely external evaluation became more likely (odds ratio 1.08/year, 95% CI 0.89–1.34). Of the 80, 45% were externally validated, 24% released a model a reader could apply, 23% assessed calibration, 16% justified a sample size, and five compared performances across patient characteristics against eight across sites; 8.8% called a random split external validation. Risk of bias was low in 16, unclear in 55 and high in nine, and none of the 18 deep-learning studies reported calibration (95% CI 0–17.6%, a loose bound on 18). We derive the design constraints these impose on a culture-independent diagnostic platform, a projection with no performance data. Validation, calibration and transparency limit this field more than accuracy does: discrimination was reported in every study appraised.

Background/Objectives: Guidance on protein and physical activity rests on population-wide minima with little differentiation by age or disease. We examined how protein intake and resistance training relate to muscle, function and mortality across clinical populations. Methods: PRISMA-guided rapid review: Fifty-eight logged PubMed queries returned 7544 unique records for screening (the analytic corpus); Europe PMC, Semantic Scholar and ClinicalTrials.gov were tested for coverage only. Estimates were extracted by a deterministic script and adjudicated by the authors; random-effects models were fitted within strata sharing an effect measure and contrast, and randomized evidence was tabulated, not pooled. Screening was single-reviewer outside a duplicate-screened sample (889 records); risk of bias was rated in duplicate. Results: In randomized trials, added protein increased strength with resistance training (+2.01%) but not without it (+0.13%); no trial of protein with resistance training reported mortality. Meeting activity guidelines was associated with lower mortality (HR 0.76, 95% CI 0.72–0.79; k = 16; I² = 64%; low certainty). Plant protein was associated with lower mortality (HR 0.76, 0.63–0.93; k = 4; very low certainty). After excluding studies at high risk of bias, the protein strata fell to single studies. The direction of protein–outcome associations differed between populations. Conclusions: Protein with resistance training is a plausible strategy for preserving muscle and function; effects on survival are untested, and population-specific protein ranges are provisional and hypothesis-generating.

Chronic non-communicable conditions – type 1 diabetes mellitus (T1DM), type 2 diabetes mellitus (T2DM), metabolic obesity syndrome (MOS), polycystic ovary syndrome (PCOS), colorectal and extra-intestinal cancers, systemic autoimmune disease, and dermatologic and gynecologic disorders linked to gut dysbiosis – share a prolonged asymptomatic phase during which conventional screening is invasive, insensitive, or resource-intensive. This review synthesizes 2021-2025 literature on fecal microbiome-based artificial intelligence (AI) diagnostics across these conditions. We surveyed machine learning classifiers trained on 16S rRNA and shotgun metagenomic data, extracting reported discriminative performance, validated microbial and short-chain fatty acid (SCFA) biomarkers, and cross-cohort reproducibility. Reported classifiers achieve areas under the curve (AUC) typically between 0.79 and 0.93 across disease domains (e.g., 0.792 for autoimmune disease subtyping, 0.82-0.90 for colorectal cancer, 0.93 for PCOS subtyping), with dietary data integration and SCFA quantification further improving discrimination. We propose a multimodal deep learning architecture – combining a microbiome transformer encoder, dietary embedding module, host feature multilayer perceptron, phylogenetic graph neural network, and cross-attention fusion layer – coupled with explainable AI (SHAP, attention heatmaps, microbial risk scores) for clinical interpretability. We conclude that fecal microbiome-based multimodal AI is a technically mature but clinically unvalidated candidate for population-scale pre-symptomatic screening, pending prospective, harmonized cross-cohort trials.

Chronic non-communicable conditions—type 1 and type 2 diabetes mellitus (T1DM, T2DM), metabolic obesity syndrome (MOS), polycystic ovary syndrome (PCOS), colorectal and extra-intestinal cancers, and systemic autoimmune disease—share a prolonged asymptomatic phase during which conventional screening is invasive, insensitive, or resource-intensive. This review synthesizes the 2021–2026 literature on fecal microbiome-based artificial intelligence (AI) diagnostics across these conditions, extracting reported discrimination, validation strategy, microbial and short-chain fatty acid (SCFA) biomarkers, and cross-cohort reproducibility. Across the primary classifier studies tabulated here, reported areas under the curve (AUCs) span 0.76–0.99 under internal validation but 0.69–0.91 under external or cross-population validation; in the four studies reporting both, the median AUC falls from 0.875 to 0.810. Verified external-validation values include 0.82 for colorectal cancer, 0.79 for T2DM and 0.792 for discrimination of systemic lupus erythematosus from rheumatoid arthritis and controls. Clinical readiness turns on this internal-to-external gap more than on the headline AUC. We propose a multimodal deep learning architecture coupled with explainable AI; no component has been implemented or evaluated on data, and it is presented as a design proposal. Fecal-microbiome-based multimodal AI is technically feasible but clinically unvalidated, pending prospective, harmonized cross-cohort trials.