deepchem

Name: deepchem
Availability: InStock
Author: leonchaox

Build machine learning pipelines for molecular property prediction from chemical datasets.

783|65|Updated Feb 27, 2026

Search Tags:#graph neural networks #smiles #deepchem #scaffold splitting #transfer learning #molecular property prediction

Ready to run

npx skills add https://github.com/LeonChaoX/qinyan-academic-skills --skill deepchem-leonchaox

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: deepchem
Source: https://github.com/LeonChaoX/qinyan-academic-skills/tree/main/skills/06-%E5%8C%96%E5%AD%A6%E4%BF%A1%E6%81%AF%E4%B8%8E%E8%8D%AF%E7%89%A9%E5%8F%91%E7%8E%B0/deepchem
Command: npx skills add https://github.com/LeonChaoX/qinyan-academic-skills --skill deepchem-leonchaox

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires deepchem, argparse, sys, numpy, and includes scripts (resource) and references (resource) components.

What problem does it solve?

DeepChem turns raw chemical and biological inputs (like SMILES, SDF files, and sequences) into machine-learning-ready datasets so you can train models for molecular property prediction and drug-discovery style tasks.

Core Features & Use Cases

Molecule data loading: Ingest CSV/SDF/FASTA data into DeepChem datasets for rapid experimentation.
Flexible featurization: Use fingerprints (e.g., ECFP/CircularFingerprint), graph features (e.g., MolGraphConvFeaturizer), descriptors (RDKitDescriptors/MordredDescriptors), and pretrained embedding-compatible inputs (Raw).
End-to-end modeling workflow: Train and evaluate traditional ML and graph neural networks with correct drug-discovery splitting via ScaffoldSplitter to reduce data leakage.
Benchmarks + transfer learning: Run MoleculeNet benchmarks and fine-tune pretrained models (e.g., ChemBERTa/GROVER/MolFormer) for smaller datasets.

Quick Start

Use the deepchem skill to train and evaluate a graph neural network on a MoleculeNet benchmark by generating scaffold-split train/valid/test datasets and fitting a model with the appropriate featurizer.

Frequently Asked Questions about deepchem

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I predict molecular properties from SMILES strings using graph neural networks?▼

To predict molecular properties from SMILES strings, you can use graph neural networks by applying featurizers like MolGraphConvFeaturizer to convert SMILES into graph features, then training DeepChem models on the resulting datasets.

What is scaffold splitting and why is it necessary for molecular property prediction?▼

Scaffold splitting is a dataset partitioning method that groups molecules by their structural scaffolds to reduce data leakage. It ensures that structurally similar molecules are kept in the same split, providing a more rigorous evaluation for molecular property prediction.

How do I apply transfer learning to a small dataset for drug discovery?▼

You can apply transfer learning to small datasets by fine-tuning pretrained molecular representation models like ChemBERTa, GROVER, or MolFormer, which leverages learned chemical features to improve prediction accuracy on limited data.

Can I use DeepChem to load and featurize raw SDF files for MoleculeNet benchmarks?▼

Yes, DeepChem supports loading raw SDF and CSV files into machine-learning-ready datasets. It provides flexible featurization options including ECFP, RDKitDescriptors, and graph features for running MoleculeNet benchmarks.

What's the best way to prevent data leakage when training models on MoleculeNet benchmarks?▼

The best way to prevent data leakage when training on MoleculeNet benchmarks is to use the ScaffoldSplitter. It partitions datasets based on molecular scaffolds, ensuring structurally distinct splits for robust model evaluation.