TorchDrug is a PyTorch-based machine learning toolkit designed for drug discovery, covering molecules, proteins, and biomedical knowledge graphs.

Does this skill support RDKit integration?

Yes, TorchDrug is fully compatible with RDKit for cheminformatics tasks and provides easy conversion between RDKit molecules and TorchDrug graph objects.

Can I use TorchDrug for protein structure prediction?

Yes, TorchDrug integrates with AlphaFold and ESM, providing models like GearNet for 3D protein structure modeling and property prediction.

What types of datasets are included in TorchDrug?

It includes over 40 curated datasets spanning molecular properties (BBBP, Tox21), protein function (GeneOntology), retrosynthesis (USPTO), and knowledge graphs (Hetionet).

Is TorchDrug suitable for de novo drug design?

Yes, it provides advanced models for molecular generation and property-guided optimization using reinforcement learning (GCPN) and graph autoregressive flows.

TorchDrug Machine Learning

Name: TorchDrug Machine Learning
Author: x-cmd

byx-cmd

•

データサイエンスとML

Facilitates machine learning for drug discovery using a PyTorch-based toolkit for molecular property prediction, protein modeling, and retrosynthesis.

概要

TorchDrug is a specialized PyTorch-powered machine learning library designed specifically for the drug discovery and molecular science domains. It enables AI agents and developers to build complex pipelines for graph-based molecular property prediction, 3D protein structure modeling, biomedical knowledge graph reasoning, and de novo molecular generation. With access to over 40 curated datasets and state-of-the-art architectures like GNNs, SchNet, and ESM, this skill streamlines the transition from raw SMILES or PDB data to actionable biological insights and chemical synthesis plans.

主な機能

Automated workflows for molecular generation and retrosynthesis planning
Access to 40+ curated datasets for drug discovery and biomedical tasks
8 GitHub stars
Biomedical knowledge graph reasoning and link prediction models
Comprehensive Graph Neural Network (GNN) suite for molecules and proteins
Protein modeling with sequence (ESM) and structure (GearNet) capabilities

ユースケース

Modeling protein functions and interactions from PDB structures or sequences
Predicting ADMET properties and toxicity for drug candidate screening
Designing novel molecular structures with specific binding affinities

概要

主な機能

Automated workflows for molecular generation and retrosynthesis planning
Access to 40+ curated datasets for drug discovery and biomedical tasks
8 GitHub stars
Biomedical knowledge graph reasoning and link prediction models
Comprehensive Graph Neural Network (GNN) suite for molecules and proteins
Protein modeling with sequence (ESM) and structure (GearNet) capabilities

ユースケース

Modeling protein functions and interactions from PDB structures or sequences
Predicting ADMET properties and toxicity for drug candidate screening
Designing novel molecular structures with specific binding affinities