Spirographic AI

Spirographic AI predicts how a molecular structure interacts with 52 biological transporters, 195 receptor sites that are organ specific, and then evaluates how the CYP 450 system breaks it down and what metabolites are formed.  This is all done from SMILES string alone. No drug name. No database record. No prior pharmacokinetic data required. 

Validation Report Transporters
Validation Report Receptors
Try Before You Buy
52

Transporter-organ pairs

92.4%

TRANSPORTER  ACCURACY

Input

SMILES STRING ONLY

8

Biological Systems

195

Receptor-organ pairs

99.4% CYP 450 Accuracy

Across 56 drugs

A 2026 peer-reviewed benchmarking study published in the Journal of Chemical Information and Modeling evaluated the leading open-access metabolite prediction models against human radiolabeled ADME data and concluded that current AI tools “remain insufficient to replace experimental studies.” The best-performing model in that study, BioTransformer 3.0, reported an F1 score of approximately 0.25 — meaning it correctly identifies fewer than half of known metabolites while generating a high rate of false positives. The Spirographic AI engine operates in a categorically different performance tier with accuracy rates of 99.4% and known limitations clearly outlined this is a fundamentaly differnt way of approaching drug development.

No More Binary Predictions for Albumin

Site I, Site II, Subdomain IB, and Subdomain IIIA binders compete for albumin occupancy differently, interact with co-administered drugs differently, and behave differently under conditions of hypoalbuminemia, renal impairment, and glycation. Knowing that a drug is highly protein-bound is useful. Knowing exactly where it binds is actionable.

Spirographic AI predicts not just whether a compound binds albumin — but where, and whether it occupies multiple sites simultaneously. With a validated dual-binding accuracy of 87.9%, the platform enables more precise pharmacological and pharmacokinetic predictions than binary protein-binding data alone can provide.

Validation Report

Transporter Systems

8 Barrier

CYP P450 and Albumin

A complete computational ADMET screening from a single SMILES string returns transporter substrate predictions across all ten systems — giving researchers the complete barrier profile for any investigational compound in one place, before wet lab work begins, saving thousands.”

Lactation

93.0%

Transporters Modeled

Placental Transfer

96.0%

Transporters Modeled

Blood Brain Barrier

94.2%

Transporters Modeled

Retinal

86.0%

Transporters Modeled

Pulmonary

91.0%

Transporters Modeled

Hepatic

91.3%

Transporters Modeled

Renal

86.9%

Transporters Modeled

Gastrointestinal

93.2%

Transporters Modeled

Albumin

86.4%

Multiple site binding predictions 

CYP 450

99.4%

7 Isoforms

Breast Milk and Placental Transfer

96.4 % and 93.7% Accuracy

Breast Milk and Placental Transfer

96.4 % and 93.7% Accuracy

No commercial platform currently offers per-transporter mechanistic prediction for either compartment. The best published academic models for placental transfer predict only binary crossing status — whether a drug crosses at all — with no information about which transporters are involved, whether they are protective efflux or exposure-increasing influx, or how transporter expression shifts across trimesters. For breast milk, the most advanced available tools predict only a bulk milk-to-plasma ratio, and the field’s own validation studies explicitly exclude known transporter substrates because existing models cannot handle them.

Spirographic AI predicts the full transporter picture the field has been routing around. Validated across 302 drugs for placental transfer (96.4% accuracy) and breast milk transfer (93.7% accuracy) — with per-transporter mechanistic output across 5 active transporters including trimester-aware OCT3 scaling, efflux vs influx classification, and fetal exposure risk stratification. This is not a binary crossing prediction. It is a complete barrier profile.