Microbe-viral co-occurrence network (adapted from James T. Morton et al., Nature Neuroscience)
Machine Learning for Translation Discovery
Feature Discovery & Prediction
Find lists of metabolites / microbes that predict outcomes (e.g. responders vs. non-responders) and hold up out-of-sample.
Responder Stratification
Classify subgroups and link them to mechanisms (pathways, taxa, enzymes).
Model Validation
Cross-validation, permutations, and test/hold-out summaries
(ROC-AUC, PR, calibration).
Typical Use Cases
Claims Support
Build compact signatures that separate treatment vs. control.
Target & Biomarker Discovery
Nominate enzymes, transporters, taxa tied to outcome.
Diagnostic Panels
Translate multi-omic signals into minimal, testable markers.
Manufacturing/QC
Detect out-of-spec lots or drift from reference profiles.
Microbiome therapeutics
Engraftment and mechanism tracking.
Deliverables
• Ranked features (metabolites/microbes) with effect sizes, direction, FDR, and pathway context
• Predictive model card: train/validation metrics, confusion matrix, permutation test and top features
• Actionable biology: pathway callouts and hypotheses (targets, biomarkers, endpoints)