ADMETlab 3.0 API & automation

Automate ADMETlab 3.0 programmatically — submit a job, poll it to completion, and retrieve predictions across its full endpoint set. Which endpoint bundle you request and whether preprocessing is enabled most affect both what comes back and how long the job takes.

API overview

The ADMETlab 3.0 API enables programmatic access to all 119 ADMET prediction endpoints. The underlying multi-task DMPNN architecture (trained on 400,000+ entries) provides consistent predictions across absorption, distribution, metabolism, excretion, and toxicity properties.

Submit jobs

• Create a job with your upload reference plus endpoint bundle (core or extended with 119 endpoints), batch size, and preprocessing toggle.
• Include a project for quota tracking and reproducibility.
• Tag runs with run_id to correlate retries or downstream analyses.
• Specify which endpoint categories to include: absorption, distribution, metabolism, excretion, toxicity, or all.

Poll status

• Poll until succeeded or failed; respect retry/backoff headers.
• Capture model_version, endpoint bundle, and preprocessing flags from metadata.

Retrieve artifacts

• Download CSV (and optional JSONL) via the artifact endpoint after success.
• Results include predictions for all requested endpoints (up to 119) with confidence scores.
• Artifacts keep your identifier columns; join on compound_id or original_index.
• Store the job link with artifacts for audit trails.
• Toxicity results include new v3.0 endpoints: nephrotoxicity, neurotoxicity, ototoxicity, and hematotoxicity.

Error handling

• Differentiate validation errors (input-specific) from service errors before retrying.
• Automatically retry only idempotent submissions; fix malformed payloads before resubmitting.
• Alert if out_of_domain or preprocessing failures exceed your acceptable threshold.
• Out-of-domain warnings indicate compounds outside the training data chemical space (400,000+ entries covering drug-like molecules).

Programmatic workflow example

1. Upload: Submit SMILES/SDF file with compound identifiers
2. Configure: Select endpoint bundle (core or all 119 endpoints)
3. Submit: Create prediction job with preprocessing enabled
4. Poll: Monitor job status until completion
5. Retrieve: Download CSV/JSONL results with all predictions
6. Process: Parse results and integrate into your pipeline