Boltz-2 structural AI · Not QSPR

Know Your Compound's Safety Risk
Before You Order the Assay

Structure-based AI screens your leads against hERG, CYP3A4, CYP2D6, and CYP2C9 safety targets — in minutes, not weeks. Filter failures early. Spend your CRO budget wisely.

Start Free — 3 Screenings View Sample Report ↓
✓ No credit card required ✓ Results in ~20 min ✓ Real Boltz-2 structural scoring ✓ Your SMILES stay private

Validation Benchmarks

Calibration run in progress — benchmarks will appear here when complete. See methodology →

Validated against PubChem-verified known inhibitors and negative controls. Hard gate: AUC ≥ 0.70 to publish.

Built for drug discovery researchers

From academic labs to early-stage biotech — anyone running hit-to-lead or lead optimization

🔬

Academic Lab

Running NSF/NIH-funded virtual screening? Free tools like pkCSM have no batch mode and no audit trail your PI can cite.

Screen your virtual library before wet-lab — with a methodology section you can drop straight into your paper.

⚗️

Early-Stage Biotech

Eurofins hERG patch clamp: $600/compound, 2-week turnaround. You have 50 leads to triage before the SAR meeting on Friday.

Filter to your top 5 candidates for $149/month — then send only those 5 to the CRO.

📋

CRO / Supplement Brand

GRAS/NDI submissions and ingredient due-diligence require defensible safety documentation. A manual lit search isn't enough anymore.

PDF-ready safety reports with full methodology. Batch-screen an entire ingredient library in hours.

How It Works

1

Submit a SMILES

Paste your compound's SMILES string or draw it in the built-in structure editor. RDKit validates and canonicalizes automatically.

2

Boltz-2 Structural Scoring

Our GPU runs Boltz-2 structure prediction — the same model behind the 2025 Passaro et al. affinity benchmark. 3 replicates per target for confidence intervals.

3

Report in Your Inbox

Download a print-ready HTML report with traffic-light risk classification, ADMET profile, Composite Toxicity Index, and full methodology.

Live Sample Reports

Real Boltz-2 predictions — not hand-tuned examples. Click a compound to load the full report.

Generating report…

Built on open science

Boltz-2

Structural AI

·

RDKit

Cheminformatics

·

ChEMBL

Benchmark data

·

PubChem

Reference compounds

Simple, transparent pricing

Monthly subscription. Cancel anytime. No hidden fees.

💡 Annual billing saves 20% — contact us

Loading plans…

Frequently asked questions

ToxScreen uses Boltz-2 structural AI calibrated against known hERG, CYP3A4, CYP2D6, and CYP2C9 inhibitors from ChEMBL and PubChem. Our validation benchmarks (AUC-ROC per target) are published on the Methodology page when calibration runs complete. On typical medicinal chemistry scaffolds, we observe AUC-ROC 0.72–0.88. ToxScreen is designed as a pre-screening filter — it meaningfully reduces false starts before wet-lab work, but does not replace IND-enabling assays.
Free tools like pkCSM and ADMETlab use QSPR (quantitative structure-property relationship) models trained on molecular fingerprints — fast but blind to 3D binding geometry. ToxScreen uses Boltz-2 structure prediction, which models actual protein-ligand binding from sequence, giving it better coverage on novel scaffolds. Additionally: ToxScreen supports batch CSV upload, generates downloadable audit-trail reports, provides confidence intervals from replicates, and offers an API for pipeline integration — none of which free tools support.
No. ToxScreen is explicitly a research and pre-screening tool. Reports cannot substitute for GLP toxicology packages, IND-enabling in vitro assays (e.g. patch-clamp hERG, CYP inhibition IC50), or any FDA/EMA/PMDA-mandated safety study. Use ToxScreen to decide which compounds deserve CRO assay time — not to replace the assays. This is stated clearly in every report.
Your SMILES are encrypted at rest and associated only with your account. We never share structures with third parties for training, marketing, or analytics. Scoring runs on our own private GPU infrastructure — no third-party AI provider sees your structures. You can request deletion of all submitted SMILES and reports at any time by emailing hello@naxiai.com.
A full-panel screening (4 targets, 3 replicates each) typically completes in 15–25 minutes. The dashboard shows live per-target progress. You can close the browser tab and return later — we'll notify you when results are ready. Queue depth varies; estimated wait time is shown at submission.

Research use only. ToxScreen is a computational pre-screening tool. Predictions are not a substitute for in vitro or in vivo toxicology studies. Decisions affecting human or animal exposure must be supported by experimentally validated assays.

Not a regulatory submission tool. ToxScreen output cannot replace IND-enabling studies, GLP toxicology packages, or any FDA / EMA / PMDA-mandated assay.

Read the full methodology →  ·  Pipeline, calibration, limitations, references, privacy, terms.