Science framework

A risk-stratified pharmacogenomic and chemometric framework.

CannaprepAI combines product chemistry, route of use, tolerance signals, and optional local-only pharmacogenomic markers into a bounded caution score. The output is not a medical dose, diagnosis, treatment plan, or promise of a specific effect. It is a conservative preparation-intensity band designed to reduce overexposure risk.

No dosing

The system maps risk to preparation intensity, not to a prescribed cannabis dose.

Optional DNA

Genetic files are processed locally in the browser and discarded after selected markers are extracted.

Conservative by design

Recognized risk factors can only hold or increase caution; they cannot make the protocol stronger.

Medical and legal boundary

This page describes an educational safety algorithm for adults 21+ where legal. Cannabis response is multifactorial and can be affected by dose, route, product chemistry, tolerance, medications, sleep, diet, stress, and setting. Users with medications, pregnancy, significant medical conditions, or prior severe adverse reactions should consult a clinician or pharmacist.

Formal model

The recommendation is a bounded caution function.

Inputs

x = (c, t, r, u, g)

c is the cannabinoid vector. t is the terpene vector. r is route of use. u is the user questionnaire. g is an optional marker-state vector.

Marker states are categorical: normal, intermediate, reduced, or unknown.

Score and bands

S(x) = S_c(c) + S_t(t) + S_r(r) + S_u(u) + S_g(g)

Every component is nonnegative. Example thresholds are tau1 = 4, tau2 = 8, and tau3 = 13.

standard
Standard preparation guidance
elevated
Gentle preparation and extra caution language
high
Minimal escalation and stronger uncertainty warnings
avoid/consult
Do not escalate; consult a qualified professional
Chemometric layer

Effective cannabinoid potential is estimated, not invented.

THC_eq = delta-9 THC + 0.877 * THCA

The 0.877 factor reflects the molecular-weight adjustment after THCA loses carbon dioxide during decarboxylation. Actual delivered exposure is lower and depends on route, temperature, time, method, and losses.

E_THC = rho(r, T, dt, M) * THC_eq

rho is an empirical delivery-efficiency factor. It should be estimated from evidence and validation data, not guessed.

Pharmacogenomic layer

Genetics can add caution, but it cannot promise an effect.

S_g(g) = sum_j w_j * phi(g_j)

phi(normal)=0, phi(intermediate)=1, phi(reduced)=2, and phi(unknown)=lambda, with 0 <= lambda <= 1.

THC-heavy products can weight CYP2C9 more strongly. CBD-heavy products can weight CYP2C19 and CYP3A caution more strongly. These are risk modifiers only, not deterministic predictors.

Provable safety behavior

The algorithm can be made conservative by construction.

Theorem 1: monotonic caution

If every component score is coordinatewise nondecreasing, adding or increasing a recognized risk feature cannot produce a less cautious recommendation.

For states x and x-prime where x-prime is at least as risky on every recognized coordinate, each component score is greater than or equal to its previous value. Their sum S(x-prime) is therefore greater than or equal to S(x). Ordered thresholds preserve that ordering, so the protocol cannot move to a less cautious band.

Theorem 2: missing genetics cannot lower caution

If unknown marker states are assigned a nonnegative uncertainty value, missing genetic data cannot reduce the score below the non-genetic baseline.

The genetic score is a weighted sum of nonnegative weights and nonnegative marker values. Unknown markers receive lambda >= 0. Therefore S_g(g) >= 0 and total score S = S_0 + S_g is never below the non-genetic score S_0.

Theorem 3: raw DNA is not required after marker extraction

For this defined scoring function, retaining raw genetic files provides no mathematical benefit after the finite marker vector has been extracted.

S_g depends only on selected marker states g. If f(raw file) = g, then S_g(raw file) = S_g(g). Any two raw files that map to the same marker vector produce the same score, so the raw file should be discarded immediately after extraction.

Algorithm

Risk-stratified preparation guidance

  1. 01Parse COA image or lab data into cannabinoid and terpene vectors.
  2. 02Compute THC_eq from delta-9 THC and THCA.
  3. 03Compute product, terpene, route, and user sensitivity scores.
  4. 04If raw DNA is provided, extract selected rsIDs locally, map them to marker states, then clear the raw file data.
  5. 05Compute total score S and map it to a risk band.
  6. 06Map the risk band to standard, gentle, minimal, or avoid/consult preparation guidance.
  7. 07Return the risk band, preparation guidance, and explanation.

Honest claims

  • Recognized risk factors cannot produce a less cautious recommendation.
  • Missing genetics cannot falsely lower caution below baseline.
  • Raw DNA storage is unnecessary for this defined algorithm.
  • COA chemistry, route, tolerance, and optional genetics can be combined into a formal risk model.

Claims not made

  • No exact ideal cannabis dose.
  • No guaranteed subjective high.
  • No claim that genotype alone predicts cannabis response.
  • No diagnosis, treatment, or therapeutic benefit claim.
Validation plan

The science roadmap separates proof from evidence.

01

COA parser accuracy

Measure correctly extracted THCA, delta-9 THC, CBD, CBDA, CBG, terpene values, batch identifiers, and testing dates.

02

Monotonic safety behavior

Generate synthetic users and verify that risk-increasing input changes never move a recommendation to a less cautious protocol.

03

Recommendation stability

Perturb uncertain COA fields and estimate how often small extraction errors change the risk band, especially near thresholds.

04

Prospective user study

Compare label-only, COA-only, COA plus questionnaire, and optional genetics flows for comprehension and overexposure outcomes.

References

Source trail for the biological claims.

FDA pharmacogenomics tables

FDA maintains pharmacogenomic biomarker labeling resources. CannaprepAI treats them as caution context, not as cannabis-specific deterministic rules.

https://www.fda.gov/drugs/science-and-research-drugs/table-pharmacogenomic-biomarkers-drug-labeling
FDA pharmacogenetic associations

FDA pharmacogenetic association resources support careful, bounded use of gene-drug evidence and discourage overextension.

https://www.fda.gov/medical-devices/precision-medicine/table-pharmacogenetic-associations
THC metabolism study, 2024

Human liver microsome work reporting CYP2C9, CYP2C19, and CYP3A contributions to cannabinoid metabolism.

https://www.sciencedirect.com/science/article/pii/S0006295224001746
Cannabis pharmacogenomics review

Review literature describing CYP2C9, CYP2C19, CYP3A4, and other pharmacogenomic signals as relevant but not individually predictive.

https://pmc.ncbi.nlm.nih.gov/articles/PMC10137111/
CDC cannabis and driving safety

CDC public health guidance on impairment, coordination, judgment, and driving risk after cannabis use.

https://www.cdc.gov/cannabis/health-effects/driving.html
CDC cannabis health effects

Public health context for cannabis risk communication, including mental health, interactions, and adverse effects.

https://www.cdc.gov/cannabis/health-effects/index.html
Investor/science summary

CannaprepAI is a formal risk-stratified chemometric guidance system. It converts cannabis lab data, terpene profiles, route of use, tolerance, and optional local pharmacogenomic markers into a bounded caution score, then maps that score to preparation-intensity bands. The biological claims remain empirical, but the algorithmic safety behavior can be proven conservative.

Open the simulatorTry local DNA mode