القائمة
Molecular ML with DeepChem - featurizers, models, and molecular property prediction
Shared operational contract for all MedSci agents: sequential execution, planning phase, retry limits, evidence standards.
Isolated exploratory code execution with medsci-sandbox tools. Use when analysis requires custom code beyond existing domain MCP tools.
AlphaFold DB for predicted protein structures and pLDDT confidence scores
Molecular biology toolkit. Use for FASTA parsing, sequence analysis, and translation.
ChEMBL database access for bioactivity data and target search
Datamol for molecular manipulation, SMILES processing, and cheminformatics
| name | deepchem |
| description | Molecular ML with DeepChem - featurizers, models, and molecular property prediction |
DeepChem is a Python library for molecular machine learning. It provides featurizers, datasets, model wrappers, and splitters for drug discovery and materials science tasks.
dc.feat.CircularFingerprint(size=1024, radius=2)dc.feat.ConvMolFeaturizer() -- converts molecules to graph objects.dc.feat.WeaveFeaturizer() -- atom and pair features.dc.feat.RDKitDescriptors() -- 200+ physicochemical descriptors.import deepchem as dc
# Load or create dataset from SMILES
featurizer = dc.feat.CircularFingerprint(size=1024, radius=2)
loader = dc.data.CSVLoader(tasks=["activity"], feature_field="smiles", featurizer=featurizer)
dataset = loader.create_dataset("data.csv")
# Split
splitter = dc.splits.ScaffoldSplitter()
train, valid, test = splitter.train_valid_test_split(dataset)
# Train model
model = dc.models.MultitaskClassifier(n_tasks=1, n_features=1024, layer_sizes=[512, 256])
model.fit(train, nb_epoch=50)
# Evaluate
metric = dc.metrics.Metric(dc.metrics.roc_auc_score)
print(model.evaluate(test, [metric]))
featurizer = dc.feat.ConvMolFeaturizer()
model = dc.models.GraphConvModel(n_tasks=1, mode="classification")
RandomSplitter, ScaffoldSplitter (preferred for generalization), ButinaSplitter.dc.molnet.load_*() (e.g., load_tox21(), load_bbbp()).pip install deepchem.