DeepChem
Deep learning for the life sciences: drug discovery, quantum chemistry, materials science, bioinformatics.
When to Use This Skill
- Building ML models on molecular datasets (SMILES, graphs, fingerprints)
- Working with MoleculeNet benchmark datasets
- Predicting molecular properties (solubility, toxicity, binding affinity)
- Protein-ligand interaction modeling
- Quantum chemistry property prediction (QM9, GDB datasets)
- Featurizing molecules for downstream ML tasks
- Virtual screening and drug discovery pipelines
Quick Start — Standard Workflow
import deepchem as dc
# 1. Load dataset with featurizer
tasks, datasets, transformers = dc.molnet.load_delaney(featurizer='GraphConv')
train_dataset, valid_dataset, test_dataset = datasets
# 2. Create model
model = dc.models.GraphConvModel(n_tasks=1, mode='regression', dropout=0.2)
# 3. Train
model.fit(train_dataset, nb_epoch=100)
# 4. Evaluate
metric = dc.metrics.Metric(dc.metrics.pearson_r2_score)
train_score = model.evaluate(train_dataset, [metric], transformers)
test_score = model.evaluate(test_dataset, [metric], transformers)
# 5. Predict
predictions = model.predict_on_batch(test_dataset.X[:10])
Router — What to Read
| Task |
Reference |
| Dataset creation, access, splitters |
references/core-concepts.md |
| Training workflow, metrics, hyperopt, multitask |
references/model-training.md |
| Fingerprints, GCN, ChemBERTa, graph models |
references/mol-machine-learning.md |
| MoleculeNet, protein-ligand, virtual screening |
references/drug-discovery.md |
| QM9, DeepQMC, materials science |
references/quantum-materials.md |
Installation
pip install --pre deepchem # with TensorFlow
pip install --pre deepchem[torch] # with PyTorch
pip install --pre deepchem[jax] # with JAX
import deepchem as dc
dc.__version__ # verify installation
Key Submodules
| Submodule |
Role |
dc.molnet |
MoleculeNet dataset loaders |
dc.models |
All model classes |
dc.feat |
Featurizers |
dc.metrics |
Evaluation metrics |
dc.splits |
Dataset splitters |
dc.data |
Dataset classes |
dc.trans |
Transformers (normalization, etc.) |
Related Skills
rdkit-patterns - Molecular manipulation before DeepChem ingestioncheminformatics - SMILES, molecular representations reference
1---2name: deepchem3description: Use when working with DeepChem for molecular machine learning, drug discovery, quantum chemistry, materials science, or bioinformatics. Handles molecular datasets, featurization strategies, model training/evaluation, and predictions on chemical data.4---56# DeepChem78Deep learning for the life sciences: drug discovery, quantum chemistry, materials science, bioinformatics.910## When to Use This Skill1112- Building ML models on molecular datasets (SMILES, graphs, fingerprints)13- Working with MoleculeNet benchmark datasets14- Predicting molecular properties (solubility, toxicity, binding affinity)15- Protein-ligand interaction modeling16- Quantum chemistry property prediction (QM9, GDB datasets)17- Featurizing molecules for downstream ML tasks18- Virtual screening and drug discovery pipelines1920## Quick Start — Standard Workflow2122```python23import deepchem as dc2425# 1. Load dataset with featurizer26tasks, datasets, transformers = dc.molnet.load_delaney(featurizer='GraphConv')27train_dataset, valid_dataset, test_dataset = datasets2829# 2. Create model30model = dc.models.GraphConvModel(n_tasks=1, mode='regression', dropout=0.2)3132# 3. Train33model.fit(train_dataset, nb_epoch=100)3435# 4. Evaluate36metric = dc.metrics.Metric(dc.metrics.pearson_r2_score)37train_score = model.evaluate(train_dataset, [metric], transformers)38test_score = model.evaluate(test_dataset, [metric], transformers)3940# 5. Predict41predictions = model.predict_on_batch(test_dataset.X[:10])42```4344## Router — What to Read4546| Task | Reference |47|------|-----------|48| Dataset creation, access, splitters | `references/core-concepts.md` |49| Training workflow, metrics, hyperopt, multitask | `references/model-training.md` |50| Fingerprints, GCN, ChemBERTa, graph models | `references/mol-machine-learning.md` |51| MoleculeNet, protein-ligand, virtual screening | `references/drug-discovery.md` |52| QM9, DeepQMC, materials science | `references/quantum-materials.md` |5354## Installation5556```bash57pip install --pre deepchem # with TensorFlow58pip install --pre deepchem[torch] # with PyTorch59pip install --pre deepchem[jax] # with JAX60```6162```python63import deepchem as dc64dc.__version__ # verify installation65```6667## Key Submodules6869| Submodule | Role |70|-----------|------|71| `dc.molnet` | MoleculeNet dataset loaders |72| `dc.models` | All model classes |73| `dc.feat` | Featurizers |74| `dc.metrics` | Evaluation metrics |75| `dc.splits` | Dataset splitters |76| `dc.data` | Dataset classes |77| `dc.trans` | Transformers (normalization, etc.) |7879## Related Skills8081- `rdkit-patterns` - Molecular manipulation before DeepChem ingestion82- `cheminformatics` - SMILES, molecular representations reference
Kdevos12