Claim

A homomorphically encrypted multi-site registry that aggregates TPMT and NUDT15 genotype-phenotype data will recover nearly the same risk ranking for thiopurine myelotoxicity as plaintext pooled analysis, while increasing site participation in privacy-sensitive autoimmune and inflammatory disease networks.

Why this matters

Genotype-guided thiopurine dosing is clinically useful, but many institutions hesitate to share pharmacogenomic data across borders or governance frameworks. DeSci infrastructure could widen validation cohorts, but only if privacy-preserving computation is good enough to retain the clinical signal.

Mechanistic rationale

• TPMT and NUDT15 variants are strong biological determinants of thiopurine toxicity.
• The relevant validation outputs are mostly aggregate statistics: odds ratios, calibration, subgroup toxicity rates, and net benefit.
• These computations should be feasible under modern multi-key homomorphic encryption workflows without exposing raw genotypes.

Testable design

1. Build matched decentralized cohorts across autoimmune disease sites using encrypted summary computation.
2. Compare encrypted vs plaintext analyses for:
   • variant effect ranking
   • AUROC / calibration for myelotoxicity prediction
   • subgroup dose-adjustment benefit
3. Measure participation rate, time-to-agreement, and governance acceptance.

Quantitative prediction

Encrypted analysis will preserve the direction and approximate magnitude of TPMT/NUDT15 toxicity effects while improving participation from sites that would decline plaintext genomic pooling.

What would falsify it

If encrypted computation materially degrades calibration, misranks the strongest pharmacogenomic predictors, or imposes impractical latency/cost, the hypothesis fails.

References

1. CPIC Guideline for Thiopurine Dosing Based on TPMT and NUDT15 Genotypes: 2025 Update. Clin Pharmacol Ther 2026. DOI: 10.1002/cpt.70209
2. TPMT in the treatment of inflammatory bowel disease with azathioprine. Gut 2003. DOI: 10.1136/gut.52.5.767
3. NUDT15 C415T variant compared with TPMT genotyping in predicting azathioprine-induced leucopenia. Aliment Pharmacol Ther 2021. DOI: 10.1111/apt.16137
4. Privacy-preserving framework for genomic computations via multi-key homomorphic encryption. Bioinformatics 2025. DOI: 10.1093/bioinformatics/btae754