{ "schema_version": 1, "generated_at": "2026-08-28T11:43:22+00:00", "experiment_count": 9, "hub": { "name": "Oralbiome", "url": "https://oralbiome.scarpelli.ai", "tagline": "Oral-microbiome research. The signal in your mouth is real. It just doesn't travel well." }, "studies": [ { "id": "cross-population-t2d", "title": "Cross-population oral-microbiome T2DM screening", "status": "active", "flagship": true, "paper": { "title": "Predicting Type 2 Diabetes from Oral Microbiome Composition", "authors": "Scarpelli & Akhter, 2026", "pdf": "/preprint.pdf" }, "key_finding": "The signal transfers within related populations but fades with distance. In South Africa it reverses: a strong within-cohort signal (CV 0.853) that the pooled model reads backwards (transfer 0.320). Not noise, not sequencing depth, a real population or study-specific reversal of the microbiome-to-diabetes relationship.", "reproduces": { "cv_auc": 0.81, "portugal_auc": 0.698, "holdout3_auc": 0.673, "targets": { "cv": 0.803, "portugal": 0.698, "holdout3": 0.673 }, "status": "pass" }, "leaderboard": [ { "cohort": "Portugal", "population": "European", "country": "Portugal", "specimen": "saliva", "kind": "geographic holdout", "auc": 0.698, "n": 50, "n_diabetic": 25, "n_control": 25, "within_auc": 0.859, "lng": -8.2, "lat": 39.4, "accession": "PRJNA679485", "ref": "Almeida-Santos et al., 2021", "inverted": false }, { "cohort": "India/Mumbai", "population": "South Asian", "country": "India", "specimen": "plaque", "kind": "new cohort", "auc": 0.621, "n": 60, "n_diabetic": 40, "n_control": 20, "within_auc": 0.638, "lng": 72.9, "lat": 19.1, "accession": "PRJNA1240053", "ref": "Bhosale et al., Diseases 2026", "inverted": false }, { "cohort": "Mexico", "population": "Latin American", "country": "Mexico", "specimen": "saliva", "kind": "new cohort", "auc": 0.564, "n": 87, "n_diabetic": 41, "n_control": 46, "within_auc": 0.566, "lng": -102.5, "lat": 23.6, "accession": "PRJNA1208116", "ref": "Martinez-Gregorio et al., J Appl Microbiol 2026", "inverted": false }, { "cohort": "South Africa", "population": "African", "country": "South Africa", "specimen": "plaque", "kind": "new cohort", "auc": 0.32, "n": 88, "n_diabetic": 57, "n_control": 31, "within_auc": 0.853, "lng": 22.9, "lat": -30.6, "accession": "PRJNA723337", "ref": "Matsha et al., J Dent Res 2020", "inverted": true } ], "marker_panel": { "cohorts": [ "NHANES (US)", "China", "Portugal", "India/Mumbai", "Mexico", "South Africa" ], "cohort_meta": [ { "name": "NHANES (US)", "population": "US", "n": 9683, "cases": 830, "controls": 8853 }, { "name": "China", "population": "Chinese", "n": 624, "cases": 510, "controls": 114 }, { "name": "Portugal", "population": "European", "n": 50, "cases": 25, "controls": 25 }, { "name": "India/Mumbai", "population": "South Asian", "n": 60, "cases": 40, "controls": 20 }, { "name": "Mexico", "population": "Latin American", "n": 87, "cases": 41, "controls": 46 }, { "name": "South Africa", "population": "African", "n": 88, "cases": 57, "controls": 31 } ], "heatmap": [ { "genus": "Peptoniphilus", "r": [ 0.017, 0.097, 0.139, 0.15, 0.099, 0.103 ], "class": "robust_up", "agreement": 1.0, "mean_r": 0.101 }, { "genus": "Enterococcus", "r": [ 0.009, 0.074, 0.138, 0.149, 0.086, 0.072 ], "class": "robust_up", "agreement": 1.0, "mean_r": 0.088 }, { "genus": "Alloscardovia", "r": [ 0.052, 0.124, 0.027, 0.149, 0.054, 0.086 ], "class": "robust_up", "agreement": 1.0, "mean_r": 0.082 }, { "genus": "Lactobacillus", "r": [ 0.067, 0.145, -0.088, 0.062, 0.195, 0.065 ], "class": "robust_up", "agreement": 0.83, "mean_r": 0.074 }, { "genus": "Streptococcus", "r": [ -0.004, 0.231, 0.136, 0.397, 0.101, -0.118 ], "class": "robust_up", "agreement": 0.8, "mean_r": 0.124 }, { "genus": "Bifidobacterium", "r": [ 0.058, 0.371, -0.078, 0.084, 0.112, -0.013 ], "class": "robust_up", "agreement": 0.8, "mean_r": 0.089 }, { "genus": "Fretibacterium", "r": [ 0.009, -0.196, -0.133, -0.134, -0.06, -0.168 ], "class": "robust_down", "agreement": 1.0, "mean_r": -0.114 }, { "genus": "Johnsonella", "r": [ -0.033, -0.144, -0.093, -0.105, -0.148, -0.061 ], "class": "robust_down", "agreement": 1.0, "mean_r": -0.098 }, { "genus": "Cloacibacterium", "r": [ -0.016, -0.2, -0.215, -0.206, -0.095, 0.074 ], "class": "robust_down", "agreement": 0.8, "mean_r": -0.11 }, { "genus": "Treponema", "r": [ 0.008, -0.17, 0.105, -0.175, -0.073, -0.25 ], "class": "robust_down", "agreement": 0.8, "mean_r": -0.092 }, { "genus": "Delftia", "r": [ 0.009, -0.085, -0.065, 0.149, -0.181, -0.338 ], "class": "robust_down", "agreement": 0.8, "mean_r": -0.085 }, { "genus": "Schlegelella", "r": [ 0.007, -0.126, -0.065, 0.149, -0.181, -0.146 ], "class": "robust_down", "agreement": 0.8, "mean_r": -0.06 }, { "genus": "Serratia", "r": [ 0.009, 0.103, -0.712, 0.151, -0.181, 0.13 ], "class": "flipper", "agreement": 0.4, "mean_r": -0.083 }, { "genus": "Faecalibacterium", "r": [ -0.023, 0.406, -0.034, -0.342, -0.013, 0.467 ], "class": "flipper", "agreement": 0.67, "mean_r": 0.077 }, { "genus": "Campylobacter", "r": [ 0.001, -0.153, 0.26, 0.031, -0.007, -0.477 ], "class": "flipper", "agreement": 0.67, "mean_r": -0.057 }, { "genus": "Lysinibacillus", "r": [ 0.008, -0.293, -0.065, 0.149, -0.181, 0.394 ], "class": "flipper", "agreement": 0.4, "mean_r": 0.002 }, { "genus": "Microbacterium", "r": [ 0.009, -0.343, -0.065, 0.149, 0.118, 0.336 ], "class": "flipper", "agreement": 0.6, "mean_r": 0.034 }, { "genus": "Burkholderia", "r": [ -0.02, -0.44, -0.065, 0.149, -0.181, 0.238 ], "class": "flipper", "agreement": 0.6, "mean_r": -0.053 }, { "genus": "Bacteroides", "r": [ -0.018, 0.467, 0.169, -0.034, -0.138, -0.207 ], "class": "flipper", "agreement": 0.5, "mean_r": 0.04 }, { "genus": "Ureaplasma", "r": [ 0.004, -0.038, -0.065, 0.149, -0.201, 0.467 ], "class": "flipper", "agreement": 0.5, "mean_r": 0.053 }, { "genus": "Paenibacillus", "r": [ 0.005, -0.12, -0.065, 0.149, -0.196, 0.467 ], "class": "flipper", "agreement": 0.4, "mean_r": 0.04 }, { "genus": "Actinomyces", "r": [ 0.009, 0.079, 0.149, -0.165, -0.124, 0.492 ], "class": "flipper", "agreement": 0.6, "mean_r": 0.073 }, { "genus": "Parabacteroides", "r": [ -0.007, 0.355, -0.065, 0.149, -0.187, 0.467 ], "class": "flipper", "agreement": 0.6, "mean_r": 0.119 }, { "genus": "Atopobium", "r": [ 0.023, -0.035, -0.065, 0.149, -0.186, 0.467 ], "class": "flipper", "agreement": 0.5, "mean_r": 0.059 }, { "genus": "Dermabacter", "r": [ 0.01, -0.028, -0.065, 0.149, -0.181, 0.467 ], "class": "flipper", "agreement": 0.5, "mean_r": 0.059 }, { "genus": "Eubacterium", "r": [ 0.009, 0.106, -0.065, 0.149, -0.181, 0.467 ], "class": "flipper", "agreement": 0.6, "mean_r": 0.081 } ], "robust_up": [ "Peptoniphilus", "Enterococcus", "Alloscardovia", "Lactobacillus", "Streptococcus", "Bifidobacterium", "Pseudoramibacter", "Acinetobacter" ], "robust_down": [ "Fretibacterium", "Johnsonella", "Cloacibacterium", "Treponema", "Delftia", "Schlegelella", "Brevundimonas" ], "flippers": [ "Serratia", "Faecalibacterium", "Campylobacter", "Lysinibacillus", "Microbacterium", "Burkholderia", "Bacteroides", "Ureaplasma", "Paenibacillus", "Actinomyces", "Parabacteroides", "Atopobium", "Dermabacter", "Eubacterium", "Victivallis", "Helicobacter", "Eggerthella", "Achromobacter", "Alistipes", "Bartonella", "Akkermansia", "Granulicatella", "Staphylococcus", "Pedobacter", "Phocaeicola", "Afipia", "Segetibacter", "Eikenella", "Bordetella", "Jeotgalicoccus", "Mobiluncus", "Arcanobacterium", "Jonquetella", "Peptococcus", "Lautropia", "Cardiobacterium", "Moryella", "Butyrivibrio", "Aggregatibacter", "Aquamicrobium", "Roseomonas", "Cellulosimicrobium", "Fusobacterium", "Alloprevotella", "Rhodobacter", "Veillonella", "Brucella", "Neisseria", "Haematobacter", "Yersinia", "Mitsuokella", "Kytococcus", "Oribacterium", "Capnocytophaga", "Arsenicicoccus", "Peptostreptococcus" ], "summary": { "n_genera_evaluated": 139, "n_robust": 15, "n_flippers": 56, "reversal_prevalence_ex_nhanes": 0.71, "reversal_tested_ex_nhanes": 80, "sa_marker_flip_fraction": 0.4, "sa_markers_tested": 10, "other_cohort_flip_fraction": { "Portugal": 0.1, "India/Mumbai": 0.4, "Mexico": 0.5 } } } } ], "roadmap": [ { "title": "Reproducible from public data only", "status": "now", "why": "The release must regenerate end-to-end from public sources (NHANES public-use genus tables + SRA cohorts) with no restricted inputs, so every result is independently verifiable." }, { "title": "Close the cross-population transfer gap", "status": "now", "why": "The core scientific problem: models trained on US+Asian cohorts fall to near-random on unseen populations. Making the oral-microbiome T2DM signal generalize across populations is this project's contribution." }, { "title": "Broaden cohort diversity continuously", "status": "next", "why": "Generalization needs the populations we lack (Africa, Middle East, Latin America, more Europe). Keep ingesting new public, T2DM-labeled oral-microbiome cohorts as they appear." }, { "title": "Living, citable preprint", "status": "next", "why": "Keep docs/Preprint.md current as results evolve, and land a public release with a Zenodo DOI so the work stays discoverable and updatable rather than frozen." }, { "title": "Calibrated saliva screening tool", "status": "later", "why": "Turn the model into a population-aware, honestly-calibrated T2DM screen (evolve the HTML calculator into a validated tool), the translational end goal." } ], "training_regions": [ { "name": "NHANES (US)", "lng": -98.6, "lat": 39.8 }, { "name": "China", "lng": 113.0, "lat": 33.0 } ] }