For Contract Research Organizations

Run a Blinded Benchmark on Your Own Compounds

We don't ask you to trust our published validation. Test ChiralCall against your own ground truth first.

Submit up to 150 chiral compounds where you already know the eutomer. Get ChiralCall's predictions back. Compare them against your ground truth using our browser-based report — your answers never leave your machine. Free for 60 days. No credit card. No commitment.

How the Blinded Benchmark Works

1

Pick your test set

Choose 20–150 chiral compounds where you already know the active enantiomer. Internal compounds from recent programs work best. Include compounds from each compound class you care about. NDA available before you submit proprietary structures.

2

Submit via batch API

Send your SMILES as a JSON array to /api/v1/batch. Each result includes the predicted eutomer, a calibrated confidence score, compound class match, and per-class accuracy with Wilson confidence intervals. All processing is stateless — your SMILES are never stored.

3

Compare with your ground truth

Open the Comparison Report page. Paste ChiralCall's predictions and your known answers. The report runs entirely in your browser — your ground truth never leaves your machine and is never sent to our servers.

4

Decide with data

The report shows overall accuracy, per-class breakdown, false-confident misses, and out-of-scope calls — all computed on your chemical matter. Share internally to make an informed go/no-go decision.

Sample Batch API Output

Here's what you'll see for each compound. This example shows three well-known drugs submitted via the batch API.

POST /api/v1/batch200 OK
{
  "results": [
    {
      "smiles": "CC(C)Cc1ccc(cc1)[C@@H](C)C(=O)O",
      "prediction": "(S)-enantiomer",
      "direction": "S",
      "confidence": 92,
      "matchType": "scaffold",
      "drugClass": "Arylpropionic acid (profen NSAID)",
      "drug_class_stats": {
        "n_validated": 44,
        "accuracy_pct": 100.0,
        "wilson_ci_lower": 92.0,
        "wilson_ci_upper": 100.0,
        "tier": "T1"
      }
    },
    {
      "smiles": "OC(=O)CC(CC(=O)O)(O)C(=O)O",
      "prediction": "out_of_scope",
      "matchType": "none",
      "drugClass": null,
      "note": "No matching compound class found"
    },
    {
      "smiles": "CC(=O)Oc1ccccc1C(=O)O",
      "prediction": "achiral",
      "matchType": "none",
      "drugClass": null,
      "note": "No stereocenters detected"
    }
  ],
  "summary": { "total": 3, "predicted": 1, "out_of_scope": 1, "achiral": 1 }
}

Predicted

Ibuprofen — correctly identified as an arylpropionic acid with (S)-eutomer. Confidence 92, backed by 44 validated compounds at 100% class accuracy.

Out of scope

Citric acid — no matching compound class. ChiralCall returns “out of scope” rather than a low-confidence guess.

Achiral

Aspirin — no stereocenters detected. Handled gracefully without wasting a prediction.

What the Pilot Includes

150 predictions over 60 days

Single or batch, via web UI or API

Batch API access

Submit up to 50 compounds per request, 5 batches per hour

API key generation

Create and manage API keys from your dashboard

Per-compound confidence scores

Calibrated 0–100 probability of correctness

Per-class Wilson CIs and LOO accuracy

Statistical evidence backing each prediction

Browser-based comparison report

Your ground truth never leaves your machine

Stateless processing

SMILES processed in memory only. Nothing stored.

Auto-converts to Free tier

After 60 days or 150 predictions, whichever comes first

Pilot Scoring Template

Track your benchmark in a spreadsheet with auto-calculated summary metrics.

Download XLSX template

Technical Validation Supplement

Per-class accuracy, Wilson CIs, tier assignments, and methodology detail.

Client IP Protection

We understand that CROs handle confidential client structures. ChiralCall is designed for this:

Stateless by default. SMILES are processed in memory and discarded after the response. No database writes, no logs, no disk persistence.
Encrypted in transit. All API communication uses TLS 1.2/1.3. Hosted on Vercel infrastructure (SOC 2 Type II attested).
No third-party model APIs. Predictions are computed by our own deterministic classifier. Your SMILES are never sent to external AI or LLM providers.
Comparison report runs locally. Your ground truth data stays in your browser. It is never transmitted to our servers.
NDA available. We’ll sign a mutual NDA before you submit proprietary structures.
DPA on request. Data Processing Agreement available for enterprise engagements.

Check coverage before you commit

Paste any SMILES into our free scope checker to see instantly whether ChiralCall covers your compound class — no account, no API key, no data stored.

Check Your Compounds

What ChiralCall does and does not do

ChiralCall predicts the historically favored eutomer direction within validated compound classes, for first-pass synthesis and assay prioritization. It does not replace experimental confirmation.

Compounds outside our compound classes receive an “out of scope” response rather than a forced prediction. We believe this is more useful than low-confidence guesses.

Current coverage spans pharmaceuticals and agrochemicals. Before signing up, you can check if your compounds are in scope — free, no account required. Or see the full list in the compound class browser.

After the Benchmark

If ChiralCall performs well on your compounds, the Founding Customer plan provides unlimited API access at $9,995/year — a rate locked for the lifetime of your account. For organizations processing hundreds of compounds per month, this is typically less than the cost of a single chiral resolution.

Enterprise options (custom SLA, dedicated support, higher rate limits) are available.

Academic institutions with .edu email addresses get unlimited predictions free, permanently.

ChiralCall predicts the historically favored eutomer within validated compound classes. It is a chirality decision layer for first-pass chiral prioritization — not a replacement for experimental confirmation. Predictions are provided for research use.