loader image

Turn microbiome↔host metabolite data into calibrated, explainable clinical insight

TMAP — Q-REVEL+ (Clinical-first, Microbiome↔Host)
TMAP — Q-REVEL+ • Microbiome↔Host
Circulating metabolites + host targets • MIT-A ▸ QC ▸ k-fold ▸ Calibration ▸ Counterfactuals ▸ Report
Systemic sampling ?

Specimen

Prefer blood. Urine/saliva optional. Fix fasting/time; keep technical duplicates. Metabolite-centric ?

Why metabolites

They capture microbial function and host cross-talk beyond 16S composition. Decision support ?

Notice

Clinical decision support; does not replace professional judgment or local regulations.
🗂️ CSV Data ?

Required schema

  • Columns: sample_id, replicate_id, batch_id, label and one column per metabolite (analyte_*).
  • label: 0/1 for clinical outcome.

Validation

Schema check, missing-rate and RSD per replicate before training.
No file selected
No file selected
k-fold ?

Validation

Stratified k-fold. Recommended k=5. Reported: mean AUC, Brier, calibration. L2 λ ?

Regularization

L2 term to reduce overfitting. Start at 0.1 and tune via validation. Platt ?

Calibration

Platt scaling fits a post-hoc logistic on logits to improve probability calibration.
🎯 MIT-A Weights ?

Prioritization matrix

Normalized weights that compose the Idealness index. Set once and freeze before analysis.
Impact
Safety
Operational
Novelty/IP
Scalability
Norm. weights: 0.45/0.25/0.15/0.10/0.05
📊 Summary ?

What appears

Internal & external metrics, valid analyte count after QC, and Idealness index.
Waiting…
0
Idealness
🧠 Log ?

Events

Pipeline steps, QC notices, applied filters and automatic corrections.
🧪 QC & Normalization ?

Core gates

RSD≤15% per analyte (replicates), missing<10% per analyte.
Batch z-score normalization mitigates instrumental variation.
If very few analytes survive, gates relax in a controlled way.
Top missing analytes
Worst RSD analytes
🤖 Modeling & Metrics ?

Instrumentation

Logistic regression (L2), k-fold CV, ROC AUC, Brier, calibration curve and equation.
External validation (if provided) reported with same metrics.
ROC curve
Calibration
Family importance ?

Explainability

Relative contribution by families: SCFAs, indoles, bile acids, TMAO, polyamines, phenolics, sulfates.
🧪 Simulated interventions ?

Counterfactuals

Apply percent shifts on selected families for a real sample and estimate the new calibrated risk.
SCFAs % Indoles % Bile acids % TMAO %
📄 Clinical report ?

Content

Summary metrics, family contributions, high-level conduct suggestions, and regulatory disclaimer.
Uses the data on this page. Requires a valid API key.
🔍 Audit & drift ?

Traceability

Model signature, parameters, row counts per dataset and execution time. Use as an audit trail.
🛡️ FTO Watch ?

Optional

Maintain a watchlist of patent terms. Purely informative — no effect on clinical flow or scoring.
Inactive. No impact on the clinical flow.