DeepChem — Deep Learning for Drug Discovery

DeepChem — Deep Learning for Drug Discovery

DeepChem democratises deep learning for chemistry, biology, and materials science. It provides curated datasets, featurisers, and pre-built deep learning models for molecular property prediction, virtual screening, ADMET prediction, retrosynthesis, protein-ligand binding, and quantum chemistry.

Key Features

• 40+ molecular featurisers: Morgan fingerprints, graph convolution, Coulomb matrices, 3D descriptors
• Model zoo: Graph Convolutional Network (GCN), AttentiveFP, MPNN, SchNet, DimeNet, Transformer-M
• Curated benchmark datasets: MoleculeNet (17 datasets: BBBP, Tox21, SIDER, ClinTox, …)
• ADMET prediction: absorption, distribution, metabolism, excretion, toxicity
• Protein-ligand binding affinity and virtual screening pipelines
• Supports PyTorch, TensorFlow, and JAX backends

Quick Start

import deepchem as dc

# Load BBBP (blood-brain barrier permeability) dataset
tasks, datasets, transformers = dc.molnet.load_bbbp(featurizer="GraphConv")
train, val, test = datasets

model = dc.models.AttentiveFPModel(n_tasks=1, mode="classification")
model.fit(train, nb_epoch=30)
metric = dc.metrics.Metric(dc.metrics.roc_auc_score)
print(model.evaluate(test, [metric], transformers))

npx ai-supply add deepchem-drug-discovery-ml

Curated mirror of the open-source DeepChem (MIT). Get it from the source.