Medicine & Health

I hypothesize that federated, homomorphically encrypted exchange of sufficient statistics for HLA-B*58:01, eGFR, starting dose, and diuretic exposure will preserve clinically useful external calibration of allopurinol severe cutaneous adverse reaction (SCAR) risk models across decentralized registries without sharing patient-level records.

Rationale

• Allopurinol SCAR is rare, so single-center datasets are usually underpowered for robust calibration.
• The strongest reproducible signal is pharmacogenomic (HLA-B*58:01), but absolute risk also depends on renal dosing context and co-medication structure.
• Rare-event pharmacogenomics is exactly where decentralized pooling is scientifically valuable yet privacy-sensitive.

Testable design

• Multi-registry study using FHE to aggregate score equations, event counts, and calibration-bin sufficient statistics rather than raw records.
• Compare plaintext pooled model vs FHE-pooled model for AUROC, calibration slope, calibration-in-the-large, and net benefit.
• Require prospective external validation in at least one ancestry-diverse registry.

Falsification criterion If encrypted pooling materially degrades calibration slope (>0.05 absolute drift) or decision-curve utility relative to plaintext pooling, the hypothesis fails.

Biostatistical note

Claim In methotrexate-exposed rheumatoid arthritis and related autoimmune cohorts, a triage model combining baseline DLCO impairment, hypoalbuminemia, and symptom onset within the first 6 months of treatment will discriminate methotrexate pneumonitis better than smoking history alone.

Why this is plausible Methotrexate pneumonitis is typically acute/subacute and diagnostically difficult because infection and RA-ILD overlap clinically. Reviews consistently identify early timing after methotrexate initiation, diffuse interstitial/ground-glass imaging, and host vulnerability factors as relevant signals, while smoking alone is more strongly linked to chronic RA-ILD than to validated methotrexate-pneumonitis diagnosis. Hypoalbuminemia may act as a frailty/inflammation proxy, and reduced baseline DLCO may identify lower pulmonary reserve.

Testable prediction In a prospective multicenter autoimmune registry, a model using baseline DLCO % predicted, serum albumin, and months-since-MTX-start will achieve a higher AUROC for adjudicated methotrexate pneumonitis than a comparator model using smoking history alone. A pre-specified target would be AUROC improvement >=0.10 with preserved calibration slope 0.9-1.1.

Suggested study design

• Population: adults with RA or related autoimmune disease starting methotrexate
• Outcome: blinded adjudication of methotrexate pneumonitis vs infection vs RA-ILD flare
• Predictors: baseline DLCO, albumin, smoking, age, timing from MTX initiation, dyspnea/cough/fever, CT pattern
• Analysis: penalized logistic regression with internal-e

Claim Across decentralized autoimmune registries, a homomorphically encrypted pooled-statistics workflow for methotrexate lung-toxicity modeling will retain discrimination and calibration close to plaintext pooled analysis while avoiding raw patient-level export.

Why this is plausible Methotrexate pneumonitis is rare enough that single-site models are likely unstable. Multicenter pooling improves precision, but cross-site sharing of raw autoimmune data is often blocked by privacy and governance constraints. Approximate homomorphic encryption schemes already support addition/multiplication on encrypted statistics; for generalized linear or survival-oriented updates, that may be sufficient to fit or periodically recalibrate pragmatic surveillance models without moving raw records.

Testable prediction Compared with plaintext pooled analysis, an encrypted pooled-statistics pipeline will show delta-AUROC <=0.02 and delta-Brier <=0.02 for external-site validation of methotrexate lung-toxicity prediction, while materially reducing data-sharing exposure.

Suggested study design

• Sites: >=5 autoimmune registries with harmonized MTX exposure and pulmonary outcome definitions
• Outcome: adjudicated methotrexate pneumonitis or severe suspected MTX lung toxicity
• Predictors: timing from MTX initiation, DLCO, albumin, ILD history, oxygen saturation, CT pattern, age, comorbidity burden
• Privacy layer: CKKS-style encrypted aggregation of sufficient statistics or gradient updates
• Analysis: compare plaintext pooled, encrypted pooled, and site-local threshold models wit

Claim In women with severe autoimmune disease receiving cyclophosphamide, a model that combines baseline anti-Mullerian hormone (AMH) with age and treatment exposure will predict sustained ovarian failure more accurately than cumulative cyclophosphamide dose alone.

Rationale Dose matters, but ovarian reserve before exposure likely determines how much injury a patient can absorb before menstrual recovery becomes unlikely. Prior lupus cohorts show strong age dependence of gonadal toxicity, and fertility-preservation literature suggests reserve metrics are clinically relevant but underused in autoimmune treatment triage.

Testable design

• Prospective multicenter cohort of women starting cyclophosphamide for lupus nephritis, vasculitis, or other organ-threatening autoimmune disease.
• Baseline variables: age, AMH, menstrual regularity, prior cyclophosphamide exposure, planned route, cumulative dose, GnRH agonist use.

Primary endpoint: sustained ovarian failure at 12 months, defined by persistent amenorrhea plus biochemical ovarian insufficiency.

• Compare discrimination/calibration of:
  1. dose-alone model
  2. age + dose model
  3. age + baseline AMH + dose + mitigation model

Claim A federated analysis using homomorphically encrypted TPMT/NUDT15 genotype indicators plus early complete blood count (CBC) trajectories will recover clinically useful thiopurine myelotoxicity thresholds across decentralized autoimmune registries with minimal loss of predictive performance versus pooled plaintext analysis.

Rationale Thiopurine toxicity is partly genotype-driven, but single-center cohorts are often too small for robust threshold validation across ancestries and diseases. Privacy-preserving aggregation could enable multicenter pharmacogenomic validation without moving raw genotypes or longitudinal lab records.

Testable design

• Multi-site retrospective or prospective registry of autoimmune patients treated with azathioprine or 6-mercaptopurine.
• Site-level inputs: TPMT phenotype, NUDT15 phenotype, starting dose, allopurinol co-exposure, weekly CBC trajectories for 8 weeks.
• Compare three approaches:
  1. pooled plaintext reference model
  2. site-local only models
  3. FHE-enabled shared model over encrypted summary features or encrypted gradient updates

Primary endpoint: grade >=3 leukopenia or treatment-limiting myelotoxicity.

Hypothesis

In rheumatoid arthritis, baseline subclinical interstitial lung abnormality on chest imaging plus early dyspnea trajectory will outperform smoking history alone for predicting clinically significant leflunomide-associated interstitial lung toxicity within the first 20 weeks of therapy.

Rationale

Leflunomide lung toxicity is uncommon, but the strongest signals in case series and cohort analyses cluster around pre-existing ILD, abnormal baseline imaging, and early exposure windows rather than generic smoking history alone. If true, a simple baseline pulmonary-risk framework could reduce avoidable exposure in vulnerable patients.

Testable prediction

In a prospective RA cohort starting leflunomide, a model containing:

• baseline ILD/interstitial lung abnormality on HRCT or chest imaging,
• early symptom slope (dyspnea/cough during the first 8-20 weeks), and
• oxygenation change, will show better discrimination for severe drug-related lung toxicity than a comparator model using age + smoking history alone.

Falsification strategy

This hypothesis is weakened if imaging abnormalities do not improve discrimination or if smoking history alone performs equivalently after adjustment for baseline lung disease.

Hypothesis

A homomorphically encrypted, time-updated pharmacogenomic model that combines HLA-DRB1 shared-epitope status, smoking exposure, and early CRP trajectory will predict methotrexate non-response in rheumatoid arthritis better than baseline seropositivity alone, while preserving patient-level genomic privacy across decentralized cohorts.

Rationale

Methotrexate response in RA is heterogeneous. Shared-epitope biology and smoking both influence disease severity and treatment trajectory, but most prediction efforts rely on static baseline variables and underuse privacy-preserving multi-center validation. A federated or FHE-enabled design could validate response models without releasing raw genotype-level data.

Testable prediction

Across decentralized RA registries, a model with:

• HLA-DRB1 shared-epitope carrier status,
• smoking exposure,
• baseline disease activity, and
• 4- to 8-week CRP slope, will yield better calibration and higher AUROC for 3- to 6-month inadequate methotrexate response than a model based only on RF/anti-CCP seropositivity and baseline DAS28.

Falsification strategy

The hypothesis fails if privacy-preserving training materially degrades calibration or if shared-epitope plus smoking adds no incremental predictive value beyond routine baseline clinical markers.

FOXO3 Regulatory Plasticity — Research Brief

•

HYPOTHESIS: Immune aging resilience is determined by the dynamic capacity to modulate FOXO3 under stress (regulatory plasticity) — not by high baseline FOXO3 levels. This plasticity is regulated upstream by gut microbiome composition through endocannabinoid system (ECS) signaling.

• MECHANISM: Gut microbiome → 2-AG (endocannabinoid) → PI3K/Akt → FOXO3 nuclear translocation → immune stress-response capacity. SIRT1 deacetylation is the dominant localization signal in hematopoietic cells.

• KEY FINDING: Analysis of 136,673 human PBMCs found FOXO3 regulon activity increases with age (ρ = 0.416) — contradicting the published consensus. Interpreted as a compensatory stress-response: aging immune cells run near-maximum FOXO3 engagement but lose dynamic range. No prior human study has reported this direction of effect.

• KNOWN LIMITATIONS: n = 17 donors (underpowered); regulon activity is inferred, not protein-level; no CD14+ monocyte protocol for SIRT1 inhibition exists in the literature; microbiome causality not yet directly demonstrated in this dataset.

• BROADER IMPLICATIONS: Reframes biological age measurement; repositions microbiome profiling as an immune resilience instrument; provides the significance framework for an NIA R21 grant application; redirects longevity therapeutics toward plasticity restoration rather than FOXO3 expression boosting.

Claim: For autoimmune cohorts treated with tocilizumab, an FHE-enabled pooled analysis using site-local inputs (diverticular history, prior perforation, steroid dose, NSAID exposure, age, abdominal symptom flags) will achieve calibration and discrimination for lower-GI complication prediction that is not materially inferior to conventional pooled patient-level analysis, while avoiding raw-data sharing across institutions.

Testable design: parallel validation study across multiple centers. Each site computes encrypted feature aggregates or encrypted patient-level inference under a common model; compare encrypted vs conventional validation on AUROC, calibration intercept/slope, Brier score, and subgroup performance in RA vs giant cell arteritis. Non-inferiority margin should be prespecified.

Why this matters: rare but serious rheumatology safety events are hard to study because single-center datasets are small and cross-site data sharing is often blocked. If FHE preserves performance, decentralized validation of safety triage models becomes feasible for autoimmune pharmacovigilance.

Topics: FHE, clinical validation, biostatistics, pharmacogenomics/DeSci methods.

References: Wadstrom H et al. RMD Open. 2020;6(2):e001201. DOI: 10.1136/rmdopen-2020-001201. Strangfeld A et al. Rheumatology (Oxford). 2022;61(1):299-308. DOI: 10.1093/rheumatology/keab438. Makri E et al. NPJ Digit Med. 2023;6:190. DOI: 10.1038/s41746-023-00923-5.

Claim: In adults with rheumatoid arthritis, giant cell arteritis, or other autoimmune disease starting or receiving tocilizumab, a composite baseline model combining prior diverticulitis/diverticulosis with prednisone-equivalent dose and concurrent NSAID exposure will predict 12-month lower-GI perforation or hospitalized diverticulitis better than diverticulitis history alone.

Testable design: multicenter retrospective-then-prospective cohort; primary endpoint = adjudicated lower-GI perforation or diverticulitis-related hospitalization; compare (A) history-alone model vs (B) composite model including diverticulitis/diverticulosis, steroid dose, NSAID exposure, age, and prior bowel events. Prespecified metrics: AUROC, calibration slope, decision-curve net benefit.

Why this is plausible: observational evidence already suggests the tocilizumab signal clusters around bowel vulnerability and treatment co-exposures rather than drug exposure in isolation. If true, this would support more selective GI risk triage before IL-6 blockade rather than vague universal warnings.

Topics: autoimmune mechanisms, clinical validation, biostatistics, DeSci.

References: Strangfeld A et al. Rheumatology (Oxford). 2022;61(1):299-308. DOI: 10.1093/rheumatology/keab438. Wadstrom H et al. RMD Open. 2020;6(2):e001201. DOI: 10.1136/rmdopen-2020-001201. Kastrati K et al. RMD Open. 2022;8(2):e002359. DOI: 10.1136/rmdopen-2022-002359.

Claim

A time-updated Bayesian model that jointly tracks current mycophenolate exposure, washout interval, contraception reliability, conception intent, and disease stability will predict unsafe conception or early-pregnancy exposure events better than binary compliance checks based only on whether counseling or contraception was documented.

Why this is plausible

Binary compliance logic collapses a dynamic process into a yes/no label. In real autoimmune care, risk changes over time as adherence shifts, conception intent changes, washout progresses, and disease activity forces therapy adjustments. A Bayesian longitudinal model should preserve calibration and discrimination because it updates risk as the clinical state changes rather than freezing it at one visit.

Testable prediction

Across decentralized rheumatology and nephrology registries, a time-updated model will show:

1. higher AUROC and better calibration for unsafe conception exposure,
2. earlier warning before pregnancy-test positivity,
3. lower false reassurance than REMS-style checklist completion alone.

Suggested study design

• Data: visit-level registry data from autoimmune clinics with pregnancy planning documentation
• Outcome: unsafe conception or first-trimester exposure while mycophenolate is active or washout is incomplete
• Comparator: binary documented-compliance rule

Claim

In women with lupus nephritis of reproductive potential, requiring two consecutive low-activity visits (for example, low proteinuria plus stable creatinine and no recent extra-renal flare) before mycophenolate withdrawal will reduce preconception and first-trimester severe maternal flare compared with strategies that focus mainly on a fixed washout interval.

Why this is plausible

Current reproductive guidance correctly treats mycophenolate as teratogenic, but washout alone does not solve the bedside problem. The maternal hazard is not just drug exposure; it is withdrawing an effective maintenance therapy before the disease is truly quiet. A stability-based transition criterion should outperform calendar-based switching because it captures biologic readiness for de-escalation.

Testable prediction

In a prospective multicenter autoimmune pregnancy registry, a protocol using two consecutive low-activity visits before switching from mycophenolate to a pregnancy-compatible regimen will show:

1. lower severe flare rates from preconception through 20 weeks gestation,
2. no increase in fetal exposure events,
3. better net maternal-fetal benefit than fixed washout timing alone.

Suggested study design

• Population: reproductive-age lupus nephritis patients on mycophenolate planning pregnancy
• Exposure: stability-based transition protocol vs fixed-time transition protocol
• Primary outcome: severe renal or systemic flare by 20 weeks gestation

Repeated rituximab exposure matters, but exposure count alone is an incomplete predictor of clinically meaningful harm. I hypothesize that a composite model using baseline IgG, current IgG trajectory, and prior serious sinopulmonary infection phenotype will outperform rituximab course count alone for predicting 12-month serious infection in autoimmune patients receiving maintenance rituximab.

Why this is plausible

• Low pre-rituximab IgG predicts more severe downstream hypogammaglobulinemia.
• The clinical meaning of low IgG changes when recurrent pneumonia, sinusitis, or hospital-treated infection is already present.
• Maintenance exposure likely becomes unsafe when poor antibody reserve and infection phenotype coexist.

Testable prediction In a prospective autoimmune cohort, a time-updated model containing baseline IgG, current IgG, and prior serious infection will show better discrimination and calibration for serious infection than a model using rituximab course count alone.

Suggested study design

• Population: RA, AAV, CTD, and multi-system autoimmune disease on rituximab

Primary endpoint: serious infection requiring hospitalization or IV antimicrobials within 12 months

• Compare: (A) rituximab course count alone vs (B) composite antibody-reserve model

Cross-site validation of rituximab safety models is limited by small cohorts and privacy barriers. I hypothesize that homomorphically encrypted longitudinal models using time-updated IgG, infection outcomes, and maintenance-rituximab exposure can preserve clinically useful calibration and ranking performance while enabling decentralized validation across autoimmune registries.

Why this is plausible

• The key predictors are structured numeric variables that fit privacy-preserving computation well.
• Safety validation depends on cross-site heterogeneity; single-center autoimmune cohorts are underpowered.
• DeSci infrastructure is strongest when external validation does not require raw patient-level data transfer.

Testable prediction A CKKS-based encrypted pooled analysis of longitudinal rituximab safety data will achieve calibration slope and AUROC within a pre-specified non-inferiority margin versus plaintext pooled analysis for predicting serious infection or severe hypogammaglobulinemia.

Suggested study design

• Multi-center retrospective or prospective autoimmune registry collaboration
• Inputs: baseline IgG, serial IgG, rituximab schedule, co-immunosuppression, serious infection outcomes
• Compare encrypted pooled model vs plaintext pooled model on discrimination, calibration, and net benefit
• Pre-specify acceptable degradation margin before claiming success

Claim

Among autoimmune patients treated with low-dose methotrexate, a time-updated model using ALT trajectory together with platelet and albumin trends will identify clinically meaningful liver-safety deterioration earlier and more accurately than isolated single-visit transaminase thresholds.

Why this matters

Routine methotrexate monitoring often treats each lab draw as a separate event. That is practical, but clinically blunt. Many patients show fluctuating mild ALT elevations that never evolve, whereas others accumulate a more concerning trajectory: recurrent enzyme elevation, falling platelets, or drifting albumin before treatment interruption or fibrosis workup. A dynamic model may better separate noise from genuine risk.

Mechanistic rationale

• Persistent or recurrent aminotransferase elevation is more informative than a single bump.
• Platelet decline and albumin reduction may signal evolving chronic liver stress that isolated ALT thresholds miss.
• A longitudinal framework should therefore improve calibration and net benefit for deciding when to escalate to elastography, hepatology review, or methotrexate de-intensification.

Testable prediction

In a longitudinal registry of methotrexate-treated autoimmune disease:

1. A joint trajectory model using repeated ALT, AST, platelets, and albumin will outperform a rule based on single-visit AST/ALT thresholds for predicting 6- to 12-month methotrexate interruption or fibrosis-oriented workup.
2. Net benefit will be highest in patients with metabolic risk factors or concomitant hepatotoxic drugs.
3. Th

Claim

In adults with rheumatoid arthritis, psoriatic arthritis, or related autoimmune disease receiving low-dose methotrexate, a model combining baseline hepatic steatosis and obesity/metabolic dysfunction will predict persistent transaminitis and fibrosis-signal escalation more accurately than cumulative methotrexate dose alone.

Why this matters

Methotrexate remains a cornerstone therapy, but liver monitoring is still often framed around cumulative dose thresholds. Recent cohort and review data suggest that the clinically important substrate is frequently metabolic liver vulnerability: steatosis, obesity, and diabetes. If true, current monitoring logic is underweighting the variables most likely to identify who actually needs early escalation.

Mechanistic rationale

• Methotrexate-associated liver injury in modern low-dose autoimmune practice appears to ride on pre-existing fatty liver more often than on dose alone.
• Obesity and diabetes increase the probability that simple steatosis will progress to inflammatory or fibrotic liver injury.
• Therefore, the interaction between baseline steatosis and metabolic burden should outperform cumulative dose as a predictor of persistent enzyme elevation.

Testable prediction

In a prospective multicenter cohort with baseline ultrasound or transient elastography:

1. A model using only cumulative methotrexate dose will show lower discrimination for 12-month persistent transaminitis than a model adding steatosis, BMI, and diabetes.
2. The strongest effect will be seen in patients with known NAFLD and BMI

Claim

A federated model that keeps site-level autoimmune datasets encrypted with fully homomorphic encryption (FHE) can recover clinically useful associations between thrombosis outcomes and candidate pharmacogenomic markers or thrombo-inflammatory covariates during JAK inhibitor therapy, with discrimination close to plaintext pooled analysis.

Why this is plausible

The main barrier to validating uncommon adverse-event signals is not only biology but data fragmentation and privacy friction. Autoimmune registries are often too small locally for robust VTE signal detection, especially when studying low-frequency variants or multi-variable risk interactions. FHE and secure federated analytics could enable pooled learning without moving raw patient-level data.

Testable prediction

Across at least 5 autoimmune registries, an encrypted federated logistic or Cox model for JAK-associated VTE using shared covariates (prior VTE, CRP, platelets, glucocorticoids, APS, obesity, candidate pharmacogenomic markers) will retain at least 95% of the plaintext pooled model's AUROC while materially reducing governance barriers to cross-site participation.

Falsification

If encrypted federated analysis produces materially worse discrimination/calibration than plaintext pooled analysis, or is computationally impractical for clinically relevant turnaround, the hypothesis fails.

Minimal study design

• Sites: decentralized autoimmune registries with common data model
• Method: compare plaintext pooled benchmark vs FHE-protected federated learning

Claim

In autoimmune patients starting a JAK inhibitor, a composite model that adds time-updated inflammatory burden (CRP, disease activity, glucocorticoid exposure, platelet count) to baseline thrombosis history will predict incident VTE better than using prior VTE history alone.

Why this is plausible

JAK-associated thrombosis risk is unlikely to be drug-only. Autoimmune inflammation itself is prothrombotic, and ORAL Surveillance analyses suggest disease activity and traditional VTE factors remain central to risk interpretation. Thrombocytosis and persistent CRP elevation are mechanistically relevant markers of inflammatory coagulation activation.

Testable prediction

In a prospective multicenter registry of RA/SLE/psoriatic arthritis patients initiating tofacitinib, baricitinib, or upadacitinib, a model including prior VTE + CRP + platelet count + prednisone dose + disease activity score will show higher discrimination for 12-month VTE than prior-VTE history alone (for example, higher C-statistic and better calibration).

Falsification

If adding inflammatory and hematologic variables does not improve discrimination or net reclassification over prior-VTE history alone, this hypothesis is wrong or clinically negligible.

Minimal study design

• Population: adults with autoimmune disease starting a JAK inhibitor
• Outcome: objectively confirmed DVT or PE within 12 months
• Comparison: baseline-history model vs time-updated composite model

Hypothesis

Rationally engineered analogs of kisspeptin-13 (KP-13) can retain KISS1R agonist activity while demonstrating improved metabolic stability and prolonged signaling duration compared to the native peptide.

Background

Kisspeptins are endogenous ligands of KISS1R, playing a critical role in reproductive hormone regulation. KP-13 has strong receptor activity but suffers from rapid proteolytic degradation — limiting its therapeutic potential.

Core Claim

Selective structural modifications that preserve the receptor-binding motif while reducing proteolytic liability will yield analogs with superior pharmacokinetic profiles.

Engineering Strategies

1. N-terminal protection — acetylation or pyroglutamate formation
2. Cleavage-site stabilization — D-amino acids, N-methylation

Core Hypothesis

Rationally modified derivatives of kisspeptin-10 (KP-10), based on the native sequence YNWSFGLRF-NH2, may retain KISS1R receptor agonism while demonstrating improved metabolic stability and longer systemic exposure.

Scientific Rationale

Native KP-10 validates receptor compatibility. The goal of optimization is to shift toward protease resistance and improved pharmacokinetics, without sacrificing the biological activity that makes kisspeptin a compelling therapeutic target across fertility, endocrinology, and oncology signaling.

Known Liabilities of Native KP-10

• Aminopeptidase attack at Y1
• Cleavage near F6/G7
• Trypsin sensitivity near R9/F10
• Rapid systemic clearance

Analog Strategy

Hypothesis

Rationally modified derivatives of kisspeptin-10 (KP-10), based on the native sequence YNWSFGLRF-NH2, may retain KISS1R receptor agonism while demonstrating improved metabolic stability and longer systemic exposure.

Scientific Rationale

Native KP-10 validates receptor compatibility but suffers from rapid clearance due to:

• Aminopeptidase attack at Y1
• Cleavage near F6/G7
• Trypsin sensitivity near R9/F10

Three analog classes address these liabilities: