STORM: a saliva-based stochastic pharmacogenomic model for rheumatology in Mexican and Indigenous patients

Synthesis after the initial hypothesis thread:

STORM is now the cleaner version of the proposal.

Core claim: A saliva-based pharmacogenomic panel, combined with a probabilistic decision-support layer, could improve prediction of methotrexate toxicity and selected adjacent rheumatology drug risks in Mexican mestizo and Indigenous patients versus European-trained priors.

Why STORM is the right frame:

• The output should be a distribution over outcomes, not a deterministic dose.
• The first clinical value is precision safety, not generic personalization.
• The model should widen uncertainty in sparse genotype cells rather than overclaim.

Minimum viable panel:

• HLA-B*58:01 for allopurinol safety
• MTHFR C677T for methotrexate toxicity calibration
• TPMT for azathioprine pathway coverage
• Optional expansion: CYP2C9 and VKORC1 for adjacent rheumatology medication context

Validation path:

1. Pre-register the panel and endpoints.
2. Run a prospective discovery cohort with saliva collection.
3. Validate calibration, toxicity prediction, discontinuation, and decision-curve utility.
4. Compare against standard clinical priors and quantify when the model is uncertain.

If this fails calibration or clinical utility in a locked prospective cohort, STORM should be falsified or redesigned.