CYP3A4 Metabolite Prediction Validation | Spirographic AI
SPIROGRAPHIC AI · VALIDATION REPORT

CYP3A4 Metabolite Prediction

Deterministic CYP3A4 metabolite structure prediction, evaluated against literature-documented metabolites and metabolism sites across a 40-drug panel.

Version 2.0 · October 5, 2026 · 40-drug panel
Validation performance

100% Top-1 Metabolite Prediction 20/20 substrates

What the engine does

Spirographic AI’s CYP3A4 Metabolite Predictor generates predicted metabolite SMILES strings and corresponding molecular structures, and identifies CYP3A4-mediated metabolism sites. Predicted metabolites are compared with literature-reported structures for exact structural matches, rather than partial or approximate matches. The system uses deterministic molecular modeling without machine learning, training data, or force-field simulation.

Compound-level evidence

Drug-level metabolite site validation

Each of the 20 CYP3A4 substrates had at least one known metabolism site ranked first. This table shows how many documented sites for each drug were identified within the top five and top ten predictions.

DrugKnown sitesFound in top 5Found in top 10
Midazolam22 / 22 / 2
Testosterone33 / 33 / 3
Nifedipine11 / 11 / 1
Diazepam22 / 22 / 2
Alprazolam21 / 22 / 2
Cocaine11 / 11 / 1
Codeine21 / 22 / 2
Lidocaine21 / 22 / 2
Omeprazole22 / 22 / 2
Tamoxifen22 / 22 / 2
Clozapine21 / 22 / 2
Lansoprazole22 / 22 / 2
Haloperidol21 / 22 / 2
Carbamazepine11 / 11 / 1
Venlafaxine21 / 22 / 2
Erythromycin11 / 11 / 1
Triazolam21 / 22 / 2
Diclofenac22 / 22 / 2
Sertraline11 / 11 / 1
Tramadol22 / 22 / 2
Total3629 / 36 · 81%36 / 36 · 100%

Top-1 identification is measured per drug (20/20). Top-5 and Top-10 recall are measured across all 36 documented metabolism sites. These are different metrics.

Validation breadth

Drug panel

The validation includes 20 CYP3A4 substrate examples spanning several therapeutic and chemical classes, plus 20 non-CYP3A4 controls. Expand either group to review the reported compounds.

20 CYP3A4 substrates · site-prediction evaluation

Midazolam, Testosterone, Nifedipine, Diazepam, Alprazolam, Cocaine, Codeine, Lidocaine, Omeprazole, Tamoxifen, Clozapine, Lansoprazole, Haloperidol, Carbamazepine, Venlafaxine, Erythromycin, Triazolam, Diclofenac, Sertraline, Tramadol

20 non-CYP3A4 controls · included in panel

Caffeine, Ibuprofen, Coumarin, Chlorzoxazone, Flurbiprofen, Melatonin, Bupropion, Alosetron, Lorazepam, Propofol, Flutamide, Selegiline, Efavirenz, Atomoxetine, Amodiaquine, Fluoxetine, Clopidogrel, Isoniazid, Ketamine, Tolbutamide

The supplied results quantify site-prediction performance on the 20 substrates. Separate classification or false-positive rates for the 20 controls were not provided.

Study scope

Validation context

This report presents the supplied CYP3A4 benchmark, including reported exact metabolite-structure matches and site-level ranking results. The site-recall table measures documented metabolism sites and is distinct from structural identity matching. Metabolite abundance and clinical outcomes were not assessed. Independent external replication is not claimed.

Source: CYP3A4 Metabolite Site Prediction — Validation Report · v2.0 · October 5, 2026 · Brandy Stoffel, MSN, RN.