DeepChem — машинное обучение на молекулах

★ 7.0 · ai-tooling

deepchem is a Claude Code skill that brings DeepChem 2.8.0 into Claude's workflow for molecular machine learning: predicting ADMET properties, toxicity, solubility, and binding affinity using traditional ML, graph neural networks, and pretrained transformer models. Three ready-to-run scripts are included — predict_solubility.py (Delaney benchmark or custom CSV), graph_neural_network.py (GCN, GAT, MPNN, AttentiveFP architectures), and transfer_learning.py (fine-tuning ChemBERTa, GROVER, MolFormer). Eight documented capability areas cover data loading, featurization with circular fingerprints and graph descriptors, scaffold splitting, model training, MoleculeNet benchmarks, transfer learning, evaluation, and inference on new SMILES. Researchers in cheminformatics and drug discovery who need rapid experimentation with diverse molecular representations will find this skill most useful; it requires Python 3.7–3.11 and RDKit as a core dependency.

• #molecular-ml
• #deep-learning
• #drug-discovery
• #admet
• #gnn
• #cheminformatics
• #transfer-learning