Tools · Custom Modeling
MPNN Models
Fast structure-property inference that highlights key molecule substructures
Hardware and price
Price per run, by step
What a typical run reserves from your wallet: the catalog's expected runtime on each piece of hardware × your plan's per-second rate. Larger inputs reserve more, and you're charged only for the seconds the job actually runs.
Saved Model Scoring
| Hardware | Typical runtime | Basic | Starter | Pro | Enterprise |
|---|---|---|---|---|---|
| CPUDefault | 5 min | $0.042 | $0.024 | $0.015 | $0.009 |
MPNN Model Training
| Hardware | Typical runtime | Basic | Starter | Pro | Enterprise |
|---|---|---|---|---|---|
| CPUDefault | 1 h | $0.50 | $0.29 | $0.18 | $0.11 |
| Nvidia T4 | 1 h | $2.05 | $1.48 | $1.19 | $0.90 |
| Nvidia L4 | 1 h | $2.81 | $2.02 | $1.58 | $1.19 |
| Nvidia A10 | 1 h | $3.85 | $2.74 | $2.20 | $1.66 |
In workflows
Chain it with other steps.
Each input and output has a data kind, so the builder only connects steps that fit.
Saved Model Scoring
Inputs
- Structures (list) · optionalCIF, PDB
- Docked poses (list) · optionalSDF
- Docked poses (list) · optionalSDF
- Table (list) · optionalCSV
- Trained model · optionalOPAQUE
- Molecules (list) · optionalSMILES, INCHI, MOLBLOCK
- Table (list) · optionalCSV
- Docked poses (list) · optionalSDF
- Table (list) · optionalCSV, TSV, JSON
- Docked poses (list) · optionalPDBQT, SDF
Outputs
- ArchiveZIP
- MoleculesSMILES
- TableCSV
- MoleculesSMILES
MPNN Model Training
Inputs
- Table (list) · optionalCSV, TSV, JSON
- Molecules (list) · optionalSMILES
Outputs
- ArchiveZIP
- Trained modelOPAQUE
Ready-made workflows that use it
- Screen a library with your target rescorerVina + GNINA + PLIP → rescore with your model → top 20Virtual screening
- Dock the analogs your own model scores bestReaction enumerate → score with your model → dock the bestLibrary design
- Train an activity model from ChEMBL data for your targetsChEMBL activities → curate → train MPNNPredictive modeling
- Train an activity model from ChEMBL data and score your library with itChEMBL activities → curate → train MPNN → score your libraryPredictive modeling
- Enumerate products and keep the reactions your model predicts bestReaction enumerate → atom-map → predict yield → keep the bestPredictive modeling
References
The method behind it.
- Esther Heid, Kevin P. Greenman, Yunsie Chung et al. (2023). Chemprop: A Machine Learning Package for Chemical Property Prediction. Journal of Chemical Information and Modeling. doi:10.1021/acs.jcim.3c01250 (opens in a new tab)
- Kevin Yang, Kyle Swanson, Wengong Jin et al. (2019). Analyzing Learned Molecular Representations for Property Prediction. Journal of Chemical Information and Modeling. doi:10.1021/acs.jcim.9b00237 (opens in a new tab)
Run via API
Submit it from your own code.
Use an API key from your account. The estimate uses your plan's rates; submitting reserves that amount from your wallet.
# Estimate the reservation for your plan (payload fields: see the API reference)
curl -X POST https://api.cognichem.com/api/v1/jobs/estimate \
-H "X-Api-Key: $COGNICHEM_API_KEY" \
-H "Content-Type: application/json" \
-d '{"job_type": "model-score", "resource": "cpu", "payload": {}}'
# Submit (retrying with the same Idempotency-Key never submits twice)
curl -X POST https://api.cognichem.com/api/v1/jobs/submit \
-H "X-Api-Key: $COGNICHEM_API_KEY" \
-H "Idempotency-Key: $(uuidgen)" \
-H "Content-Type: application/json" \
-d '{"job_name": "model-score-1", "job_type": "model-score", "resource": "cpu", "payload": { ... }}'Custom Modeling
Related tools
- QSAR Models
Compare random forest, LightGBM, Gaussian process and other QSAR models on your data and keep the best
- Target Rescorer
Learn a target-specific score from Vina, GNINA, Boltz-2, MM-GBSA and PLIP signals on your measured compounds, and see which signals predict activity
- Active Learning
Pick the next batch of compounds to make from your measurements, balancing predicted value, uncertainty and diversity
- Reaction Models
Predict reaction yield, selectivity and rates from atom-mapped reactions and their conditions
- Thermo Property Models
Correct Joback and Hansen estimates with your own measurements and predict solubility and mixture properties
- Python Script Sandbox
Short RDKit, pandas, NumPy and SciPy analyses of your results, run in isolation with no network access