torchdrug

Build and train graph neural network models on molecular and protein data.

94|11|Updated Mar 26, 2026

Search Tags:#knowledge-graph #drug-discovery #retrosynthesis #protein-modeling #gnn #torchdrug

Ready to run

npx skills add https://github.com/swaruplab/operon --skill torchdrug-swaruplab

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: torchdrug
Source: https://github.com/swaruplab/operon/tree/main/src-tauri/protocols/torchdrug
Command: npx skills add https://github.com/swaruplab/operon --skill torchdrug-swaruplab

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

TorchDrug enables researchers to rapidly build, train, and deploy graph neural network models for molecular and protein data, supporting property prediction, molecule generation, and knowledge-graph reasoning in a single framework.

Core Features & Use Cases

Modular graph neural network architectures (GIN, GCN, GAT, SchNet, GearNet) for molecules and proteins.
End-to-end pipelines with datasets, tasks, training loops, evaluation metrics, and reproducibility tools.
Support for drug discovery workflows (property prediction, retrosynthesis, and molecule generation) and protein modeling.
Knowledge-graph integration for biomedical reasoning and seamless data interop with RDKit and structure predictors like AlphaFold/ESM.

Quick Start

Load a molecular dataset and train a graph-based property prediction task using a GIN backbone.

Frequently Asked Questions about torchdrug

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

FAQPage Schema

How do I train graph neural networks for molecular property prediction?▼

Train graph neural networks for molecular property prediction by loading a molecular dataset and using a modular GIN or GCN backbone within an end-to-end pipeline that includes training loops and evaluation metrics.

Can I use graph neural networks for protein modeling and knowledge-graph reasoning?▼

Graph neural networks support protein modeling and knowledge-graph reasoning by providing architectures like GearNet to reason over biomedical knowledge graphs and predict protein properties.

What is the best way to build a drug discovery pipeline for retrosynthesis and molecule generation?▼

Build a drug discovery pipeline for retrosynthesis and molecule generation using a modular API with pre-built architectures, dataset access, task wrappers, and reproducibility tools to enable reproducible research.

Does this graph neural network framework interoperate with RDKit and AlphaFold?▼

The framework enables seamless data interoperability with RDKit for molecular structures and integrates with structure predictors like AlphaFold and ESM for protein data processing.

What graph neural network architectures are available for molecule and protein data?▼

Available graph neural network architectures include GIN, GCN, GAT, SchNet, and GearNet, specifically designed for processing molecular and protein graph structures.

Why use a single framework for both molecule generation and knowledge-graph reasoning?▼

Using a single framework for molecule generation and knowledge-graph reasoning allows researchers to rapidly build, train, and deploy models using a modular API with shared datasets, tasks, and evaluation metrics.